faster rcnn源码理解

红太狼 2022-07-15 12:21 155阅读 0赞

理解faster rcnn的源码有几个关键点

1.算法原理、网络结构、训练过程这是基本

2.要弄懂源码里训练数据数据是怎么组织起来的，imdb，roidb，blob很关键，弄清它们的数据结构以及各个阶段是如何产生的

3.一定的python、numpy基础知识

rpn\_train.pt

#stage 1训练RPN时用的网络结构
    name: "ZF"
    layer {
      name: 'input-data'
      type: 'Python'
      top: 'data'
      top: 'im_info'
      top: 'gt_boxes'
      python_param {
        module: 'roi_data_layer.layer'#对应lib/roi_data_layer/layer.py
    #为训练RPN时为网络输入roi，此时为gt box
        layer: 'RoIDataLayer'
        param_str: "'num_classes': 21"
      }
    }
    
    #前面是ZF网，提取特征用，各个阶段共享
    #========= conv1-conv5 ============
    
    layer {
    	name: "conv1"
    	type: "Convolution"
    	bottom: "data"
    	top: "conv1"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 96
    		kernel_size: 7
    		pad: 3
    		stride: 2
    	}
    }
    layer {
    	name: "relu1"
    	type: "ReLU"
    	bottom: "conv1"
    	top: "conv1"
    }
    layer {
    	name: "norm1"
    	type: "LRN"
    	bottom: "conv1"
    	top: "norm1"
    	lrn_param {
    		local_size: 3
    		alpha: 0.00005
    		beta: 0.75
    		norm_region: WITHIN_CHANNEL
        engine: CAFFE
    	}
    }
    layer {
    	name: "pool1"
    	type: "Pooling"
    	bottom: "norm1"
    	top: "pool1"
    	pooling_param {
    		kernel_size: 3
    		stride: 2
    		pad: 1
    		pool: MAX
    	}
    }
    layer {
    	name: "conv2"
    	type: "Convolution"
    	bottom: "pool1"
    	top: "conv2"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 256
    		kernel_size: 5
    		pad: 2
    		stride: 2
    	}
    }
    layer {
    	name: "relu2"
    	type: "ReLU"
    	bottom: "conv2"
    	top: "conv2"
    }
    layer {
    	name: "norm2"
    	type: "LRN"
    	bottom: "conv2"
    	top: "norm2"
    	lrn_param {
    		local_size: 3
    		alpha: 0.00005
    		beta: 0.75
    		norm_region: WITHIN_CHANNEL
        engine: CAFFE
    	}
    }
    layer {
    	name: "pool2"
    	type: "Pooling"
    	bottom: "norm2"
    	top: "pool2"
    	pooling_param {
    		kernel_size: 3
    		stride: 2
    		pad: 1
    		pool: MAX
    	}
    }
    layer {
    	name: "conv3"
    	type: "Convolution"
    	bottom: "pool2"
    	top: "conv3"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 384
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu3"
    	type: "ReLU"
    	bottom: "conv3"
    	top: "conv3"
    }
    layer {
    	name: "conv4"
    	type: "Convolution"
    	bottom: "conv3"
    	top: "conv4"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 384
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu4"
    	type: "ReLU"
    	bottom: "conv4"
    	top: "conv4"
    }
    layer {
    	name: "conv5"
    	type: "Convolution"
    	bottom: "conv4"
    	top: "conv5"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 256
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu5"
    	type: "ReLU"
    	bottom: "conv5"
    	top: "conv5"
    }
    
    #========= RPN ============
    
    layer {
      name: "rpn_conv1"
      type: "Convolution"
      bottom: "conv5"
      top: "rpn_conv1"
      param { lr_mult: 1.0 }
      param { lr_mult: 2.0 }
      convolution_param {
        num_output: 256
        kernel_size: 3 pad: 1 stride: 1
        weight_filler { type: "gaussian" std: 0.01 }
        bias_filler { type: "constant" value: 0 }
      }
    }
    layer {
      name: "rpn_relu1"
      type: "ReLU"
      bottom: "rpn_conv1"
      top: "rpn_conv1"
    }
    layer {
      name: "rpn_cls_score"
      type: "Convolution"
      bottom: "rpn_conv1"
      top: "rpn_cls_score"
      param { lr_mult: 1.0 }
      param { lr_mult: 2.0 }
      convolution_param {
        num_output: 18   # 2(bg/fg) * 9(anchors)
        kernel_size: 1 pad: 0 stride: 1
        weight_filler { type: "gaussian" std: 0.01 }
        bias_filler { type: "constant" value: 0 }
      }
    }
    layer {
      name: "rpn_bbox_pred"
      type: "Convolution"
      bottom: "rpn_conv1"
      top: "rpn_bbox_pred"
      param { lr_mult: 1.0 }
      param { lr_mult: 2.0 }
      convolution_param {
        num_output: 36   # 4 * 9(anchors)
        kernel_size: 1 pad: 0 stride: 1
        weight_filler { type: "gaussian" std: 0.01 }
        bias_filler { type: "constant" value: 0 }
      }
    }
    layer {
       bottom: "rpn_cls_score"
       top: "rpn_cls_score_reshape"
       name: "rpn_cls_score_reshape"
       type: "Reshape"
       reshape_param { shape { dim: 0 dim: 2 dim: -1 dim: 0 } }
    }
    layer {
      name: 'rpn-data'
      type: 'Python'
      bottom: 'rpn_cls_score'
      bottom: 'gt_boxes'
      bottom: 'im_info'
      bottom: 'data'
      top: 'rpn_labels'
      top: 'rpn_bbox_targets'
      top: 'rpn_bbox_inside_weights'
      top: 'rpn_bbox_outside_weights'
      python_param {
        module: 'rpn.anchor_target_layer'#对应文件lib/rpn/anchor_target_layer.py
    #用于在原图上产生anchor，结合gt box训练rpn做box cls和box reg
        layer: 'AnchorTargetLayer'
        param_str: "'feat_stride': 16"
      }
    }
    layer {
      name: "rpn_loss_cls"
      type: "SoftmaxWithLoss"
      bottom: "rpn_cls_score_reshape"
      bottom: "rpn_labels"
      propagate_down: 1
      propagate_down: 0
      top: "rpn_cls_loss"
      loss_weight: 1
      loss_param {
        ignore_label: -1
        normalize: true
      }
    }
    layer {
      name: "rpn_loss_bbox"
      type: "SmoothL1Loss"
      bottom: "rpn_bbox_pred"
      bottom: "rpn_bbox_targets"
      bottom: "rpn_bbox_inside_weights"
      bottom: "rpn_bbox_outside_weights"
      top: "rpn_loss_bbox"
      loss_weight: 1
      smooth_l1_loss_param { sigma: 3.0 }
    }
    
    #========= RCNN ============
    # Dummy layers so that initial parameters are saved into the output net
    
    layer {
      name: "dummy_roi_pool_conv5"
      type: "DummyData"
      top: "dummy_roi_pool_conv5"
      dummy_data_param {
        shape { dim: 1 dim: 9216 }
        data_filler { type: "gaussian" std: 0.01 }
      }
    }
    layer {
      name: "fc6"
      type: "InnerProduct"
      bottom: "dummy_roi_pool_conv5"
      top: "fc6"
      param { lr_mult: 0 decay_mult: 0 }
      param { lr_mult: 0 decay_mult: 0 }
      inner_product_param {
        num_output: 4096
      }
    }
    layer {
      name: "relu6"
      type: "ReLU"
      bottom: "fc6"
      top: "fc6"
    }
    layer {
      name: "fc7"
      type: "InnerProduct"
      bottom: "fc6"
      top: "fc7"
      param { lr_mult: 0 decay_mult: 0 }
      param { lr_mult: 0 decay_mult: 0 }
      inner_product_param {
        num_output: 4096
      }
    }
    layer {
      name: "silence_fc7"
      type: "Silence"
      bottom: "fc7"
    }

上面需要注意的是rpn\_cls\_score层为每个位置的9个anchor做的只是bg/fg的二分类，而不管具体是fg的话属于那一类别，rpn阶段完成这个任务就够了，后面fast rcnn可以对region proposal进行细分和位置精修

roi\_data\_layer/layer.py

#coding:utf-8
    # --------------------------------------------------------
    # Fast R-CNN
    # Copyright (c) 2015 Microsoft
    # Licensed under The MIT License [see LICENSE for details]
    # Written by Ross Girshick
    # --------------------------------------------------------
    
    """The data layer used during training to train a Fast R-CNN network.
    
    RoIDataLayer implements a Caffe Python layer.
    """
    
    import caffe
    from fast_rcnn.config import cfg
    from roi_data_layer.minibatch import get_minibatch
    import numpy as np
    import yaml
    from multiprocessing import Process, Queue
    
    
    #为网络输入roi
    class RoIDataLayer(caffe.Layer):
        """Fast R-CNN data layer used for training."""
    
        def _shuffle_roidb_inds(self):
            """Randomly permute the training roidb."""
            if cfg.TRAIN.ASPECT_GROUPING:
                widths = np.array([r['width'] for r in self._roidb])
                heights = np.array([r['height'] for r in self._roidb])
                horz = (widths >= heights)
                vert = np.logical_not(horz)
                horz_inds = np.where(horz)[0]
                vert_inds = np.where(vert)[0]
                inds = np.hstack((
                    np.random.permutation(horz_inds),
                    np.random.permutation(vert_inds)))
                inds = np.reshape(inds, (-1, 2))
                row_perm = np.random.permutation(np.arange(inds.shape[0]))
                inds = np.reshape(inds[row_perm, :], (-1,))
                self._perm = inds
            else:
                self._perm = np.random.permutation(np.arange(len(self._roidb)))
            self._cur = 0
    #得到下一个batch训练用的图像的index，默认一次两张图片
        def _get_next_minibatch_inds(self):
            """Return the roidb indices for the next minibatch."""
    #如果所有图片都用完了，打乱顺序，roidb由每张图片的rois集合构成
            if self._cur + cfg.TRAIN.IMS_PER_BATCH >= len(self._roidb):
                self._shuffle_roidb_inds()
    #从_cur记录的位置开始选择cfg.TRAIN.IMS_PER_BATCH张图片作为训练用
            db_inds = self._perm[self._cur:self._cur + cfg.TRAIN.IMS_PER_BATCH]
            self._cur += cfg.TRAIN.IMS_PER_BATCH
            return db_inds
    #取得训练用的blob
        def _get_next_minibatch(self):
            """Return the blobs to be used for the next minibatch.
    
            If cfg.TRAIN.USE_PREFETCH is True, then blobs will be computed in a
            separate process and made available through self._blob_queue.
            """
            if cfg.TRAIN.USE_PREFETCH:
                return self._blob_queue.get()
            else:
                db_inds = self._get_next_minibatch_inds()
                minibatch_db = [self._roidb[i] for i in db_inds]
    #函数在lib/roi_data_layer/minibatch.py里实现
                return get_minibatch(minibatch_db, self._num_classes)
    
        def set_roidb(self, roidb):
            """Set the roidb to be used by this layer during training."""
            self._roidb = roidb
            self._shuffle_roidb_inds()
            if cfg.TRAIN.USE_PREFETCH:
                self._blob_queue = Queue(10)
                self._prefetch_process = BlobFetcher(self._blob_queue,
                                                     self._roidb,
                                                     self._num_classes)
                self._prefetch_process.start()
                # Terminate the child process when the parent exists
                def cleanup():
                    print 'Terminating BlobFetcher'
                    self._prefetch_process.terminate()
                    self._prefetch_process.join()
                import atexit
                atexit.register(cleanup)
    #该层初始化时调用
        def setup(self, bottom, top):
            """Setup the RoIDataLayer."""
    
            # parse the layer parameter string, which must be valid YAML
            layer_params = yaml.load(self.param_str_)
    
            self._num_classes = layer_params['num_classes']
    
            self._name_to_top_map = {}
    
            # data blob: holds a batch of N images, each with 3 channels
            idx = 0
            top[idx].reshape(cfg.TRAIN.IMS_PER_BATCH, 3,
                max(cfg.TRAIN.SCALES), cfg.TRAIN.MAX_SIZE)
            self._name_to_top_map['data'] = idx
            idx += 1
    #如果要训练RPN网，roi是gt box
            if cfg.TRAIN.HAS_RPN:
                top[idx].reshape(1, 3)
                self._name_to_top_map['im_info'] = idx
                idx += 1
    
                top[idx].reshape(1, 4)
                self._name_to_top_map['gt_boxes'] = idx
                idx += 1
    #如果是训练fast rcnn则roi是之前RPN提取的region proposal
            else: # not using RPN
                # rois blob: holds R regions of interest, each is a 5-tuple
                # (n, x1, y1, x2, y2) specifying an image batch index n and a
                # rectangle (x1, y1, x2, y2)
                top[idx].reshape(1, 5)
                self._name_to_top_map['rois'] = idx
                idx += 1
    
                # labels blob: R categorical labels in [0, ..., K] for K foreground
                # classes plus background
                top[idx].reshape(1)
                self._name_to_top_map['labels'] = idx
                idx += 1
    
                if cfg.TRAIN.BBOX_REG:
                    # bbox_targets blob: R bounding-box regression targets with 4
                    # targets per class
                    top[idx].reshape(1, self._num_classes * 4)
                    self._name_to_top_map['bbox_targets'] = idx
                    idx += 1
    
                    # bbox_inside_weights blob: At most 4 targets per roi are active;
                    # thisbinary vector sepcifies the subset of active targets
                    top[idx].reshape(1, self._num_classes * 4)
                    self._name_to_top_map['bbox_inside_weights'] = idx
                    idx += 1
    
                    top[idx].reshape(1, self._num_classes * 4)
                    self._name_to_top_map['bbox_outside_weights'] = idx
                    idx += 1
    
            print 'RoiDataLayer: name_to_top:', self._name_to_top_map
            assert len(top) == len(self._name_to_top_map)
    #作为输入前向计算
        def forward(self, bottom, top):
            """Get blobs and copy them into this layer's top blob vector."""
            blobs = self._get_next_minibatch()
    
            for blob_name, blob in blobs.iteritems():
                top_ind = self._name_to_top_map[blob_name]
                # Reshape net's input blobs
                top[top_ind].reshape(*(blob.shape))
                # Copy data into net's input blobs
                top[top_ind].data[...] = blob.astype(np.float32, copy=False)
    #不用反向传播
        def backward(self, top, propagate_down, bottom):
            """This layer does not propagate gradients."""
            pass
    
        def reshape(self, bottom, top):
            """Reshaping happens during the call to forward."""
            pass
    
    class BlobFetcher(Process):
        """Experimental class for prefetching blobs in a separate process."""
        def __init__(self, queue, roidb, num_classes):
            super(BlobFetcher, self).__init__()
            self._queue = queue
            self._roidb = roidb
            self._num_classes = num_classes
            self._perm = None
            self._cur = 0
            self._shuffle_roidb_inds()
            # fix the random seed for reproducibility
            np.random.seed(cfg.RNG_SEED)
    
        def _shuffle_roidb_inds(self):
            """Randomly permute the training roidb."""
            # TODO(rbg): remove duplicated code
            self._perm = np.random.permutation(np.arange(len(self._roidb)))
            self._cur = 0
    
        def _get_next_minibatch_inds(self):
            """Return the roidb indices for the next minibatch."""
            # TODO(rbg): remove duplicated code
            if self._cur + cfg.TRAIN.IMS_PER_BATCH >= len(self._roidb):
                self._shuffle_roidb_inds()
    
            db_inds = self._perm[self._cur:self._cur + cfg.TRAIN.IMS_PER_BATCH]
            self._cur += cfg.TRAIN.IMS_PER_BATCH
            return db_inds
    
        def run(self):
            print 'BlobFetcher started'
            while True:
                db_inds = self._get_next_minibatch_inds()
                minibatch_db = [self._roidb[i] for i in db_inds]
                blobs = get_minibatch(minibatch_db, self._num_classes)
                self._queue.put(blobs)

其中用到了lib/roi\_data\_layer/minibatch.py里的函数getminibatch

#coding:utf-8
    # --------------------------------------------------------
    # Fast R-CNN
    # Copyright (c) 2015 Microsoft
    # Licensed under The MIT License [see LICENSE for details]
    # Written by Ross Girshick
    # --------------------------------------------------------
    
    """Compute minibatch blobs for training a Fast R-CNN network."""
    
    import numpy as np
    import numpy.random as npr
    import cv2
    from fast_rcnn.config import cfg
    from utils.blob import prep_im_for_blob, im_list_to_blob
    
    
    
    #采样产生训练用的rois的blob，可以直接作为caffe的输入
    def get_minibatch(roidb, num_classes):
        """Given a roidb, construct a minibatch sampled from it."""
        num_images = len(roidb)
    #从预设的训练尺度里随机抽样用作此次产生的batch里用的roi的尺度
        # Sample random scales to use for each image in this batch
        random_scale_inds = npr.randint(0, high=len(cfg.TRAIN.SCALES),
                                        size=num_images)
    #BATCH_SIZE为一个minibatch里训练用的roi的数量
        assert(cfg.TRAIN.BATCH_SIZE % num_images == 0), \
            'num_images ({}) must divide BATCH_SIZE ({})'. \
            format(num_images, cfg.TRAIN.BATCH_SIZE)
    #每张图片上应该抽样得到的roi的数量
        rois_per_image = cfg.TRAIN.BATCH_SIZE / num_images
    #前景roi的数量
        fg_rois_per_image = np.round(cfg.TRAIN.FG_FRACTION * rois_per_image)
    #产生caffe能用的blob
        # Get the input image blob, formatted for caffe
    #_get_image_blob的实现在本文件的后面
        im_blob, im_scales = _get_image_blob(roidb, random_scale_inds)
    
        blobs = {'data': im_blob}
    #训练RPN时
        if cfg.TRAIN.HAS_RPN:
            assert len(im_scales) == 1, "Single batch only"
            assert len(roidb) == 1, "Single batch only"
            # gt boxes: (x1, y1, x2, y2, cls)
    #属于前景的roi的真实类别
            gt_inds = np.where(roidb[0]['gt_classes'] != 0)[0]
            gt_boxes = np.empty((len(gt_inds), 5), dtype=np.float32)
    #gt_boxes[i]类似于(x1,y1,x2,y2,cls)
            gt_boxes[:, 0:4] = roidb[0]['boxes'][gt_inds, :] * im_scales[0]
            gt_boxes[:, 4] = roidb[0]['gt_classes'][gt_inds]
            blobs['gt_boxes'] = gt_boxes
            blobs['im_info'] = np.array(
                [[im_blob.shape[2], im_blob.shape[3], im_scales[0]]],
                dtype=np.float32)
    #训练fast rcnn时
        else: # not using RPN
            # Now, build the region of interest and label blobs
            rois_blob = np.zeros((0, 5), dtype=np.float32)
            labels_blob = np.zeros((0), dtype=np.float32)
            bbox_targets_blob = np.zeros((0, 4 * num_classes), dtype=np.float32)
            bbox_inside_blob = np.zeros(bbox_targets_blob.shape, dtype=np.float32)
            # all_overlaps = []
            for im_i in xrange(num_images):
    #_sample_rois实现在下面，实现从每张图片的rois里采样
                labels, overlaps, im_rois, bbox_targets, bbox_inside_weights \
                    = _sample_rois(roidb[im_i], fg_rois_per_image, rois_per_image,
                                   num_classes)
    
                # Add to RoIs blob
                rois = _project_im_rois(im_rois, im_scales[im_i])
                batch_ind = im_i * np.ones((rois.shape[0], 1))
                rois_blob_this_image = np.hstack((batch_ind, rois))
                rois_blob = np.vstack((rois_blob, rois_blob_this_image))
    
                # Add to labels, bbox targets, and bbox loss blobs
                labels_blob = np.hstack((labels_blob, labels))
                bbox_targets_blob = np.vstack((bbox_targets_blob, bbox_targets))
                bbox_inside_blob = np.vstack((bbox_inside_blob, bbox_inside_weights))
                # all_overlaps = np.hstack((all_overlaps, overlaps))
    
            # For debug visualizations
            # _vis_minibatch(im_blob, rois_blob, labels_blob, all_overlaps)
    
            blobs['rois'] = rois_blob
            blobs['labels'] = labels_blob
    
            if cfg.TRAIN.BBOX_REG:
                blobs['bbox_targets'] = bbox_targets_blob
                blobs['bbox_inside_weights'] = bbox_inside_blob
                blobs['bbox_outside_weights'] = \
                    np.array(bbox_inside_blob > 0).astype(np.float32)
    
        return blobs
    
    #从一张图片的rois里采样得到roi
    def _sample_rois(roidb, fg_rois_per_image, rois_per_image, num_classes):
        """Generate a random sample of RoIs comprising foreground and background
        examples.
        """
        # label = class RoI has max overlap with
        labels = roidb['max_classes']
        overlaps = roidb['max_overlaps']
        rois = roidb['boxes']
    
        # Select foreground RoIs as those with >= FG_THRESH overlap
        fg_inds = np.where(overlaps >= cfg.TRAIN.FG_THRESH)[0]
        # Guard against the case when an image has fewer than fg_rois_per_image
        # foreground RoIs
    #fg_rois_per_this_image取fg_rois_per_this_image和fg_inds.size的较小的一个
        fg_rois_per_this_image = np.minimum(fg_rois_per_image, fg_inds.size)
        # Sample foreground regions without replacement
        if fg_inds.size > 0:
            fg_inds = npr.choice(
                    fg_inds, size=fg_rois_per_this_image, replace=False)
    
        # Select background RoIs as those within [BG_THRESH_LO, BG_THRESH_HI)
        bg_inds = np.where((overlaps < cfg.TRAIN.BG_THRESH_HI) &
                           (overlaps >= cfg.TRAIN.BG_THRESH_LO))[0]
        # Compute number of background RoIs to take from this image (guarding
        # against there being fewer than desired)
        bg_rois_per_this_image = rois_per_image - fg_rois_per_this_image
        bg_rois_per_this_image = np.minimum(bg_rois_per_this_image,
                                            bg_inds.size)
    #这里如果正负样本数量相差太大会出问题，此时应该做正负样本平衡，这里没有做
        # Sample foreground regions without replacement
        if bg_inds.size > 0:
            bg_inds = npr.choice(
                    bg_inds, size=bg_rois_per_this_image, replace=False)
    
        # The indices that we're selecting (both fg and bg)
        keep_inds = np.append(fg_inds, bg_inds)
        # Select sampled values from various arrays:
        labels = labels[keep_inds]
        # Clamp labels for the background RoIs to 0
    #设定背景roi的label为0
        labels[fg_rois_per_this_image:] = 0
        overlaps = overlaps[keep_inds]
        rois = rois[keep_inds]
    
        bbox_targets, bbox_inside_weights = _get_bbox_regression_labels(
                roidb['bbox_targets'][keep_inds, :], num_classes)
    
        return labels, overlaps, rois, bbox_targets, bbox_inside_weights
    
    def _get_image_blob(roidb, scale_inds):
        """Builds an input blob from the images in the roidb at the specified
        scales.
        """
        num_images = len(roidb)
        processed_ims = []
        im_scales = []
        for i in xrange(num_images):
    #读取roi所在的图像
            im = cv2.imread(roidb[i]['image'])
    #判断该roi是否是由水平翻转得到的
            if roidb[i]['flipped']:
    #实现水平翻转
                im = im[:, ::-1, :]
    #得到尺度
            target_size = cfg.TRAIN.SCALES[scale_inds[i]]
            im, im_scale = prep_im_for_blob(im, cfg.PIXEL_MEANS, target_size,
                                            cfg.TRAIN.MAX_SIZE)
            im_scales.append(im_scale)
            processed_ims.append(im)
    #在lib/util/blob.py里实现
        # Create a blob to hold the input images
        blob = im_list_to_blob(processed_ims)
    
        return blob, im_scales
    
    def _project_im_rois(im_rois, im_scale_factor):
        """Project image RoIs into the rescaled training image."""
        rois = im_rois * im_scale_factor
        return rois
    
    def _get_bbox_regression_labels(bbox_target_data, num_classes):
        """Bounding-box regression targets are stored in a compact form in the
        roidb.
    
        This function expands those targets into the 4-of-4*K representation used
        by the network (i.e. only one class has non-zero targets). The loss weights
        are similarly expanded.
    
        Returns:
            bbox_target_data (ndarray): N x 4K blob of regression targets
            bbox_inside_weights (ndarray): N x 4K blob of loss weights
        """
        clss = bbox_target_data[:, 0]
        bbox_targets = np.zeros((clss.size, 4 * num_classes), dtype=np.float32)
        bbox_inside_weights = np.zeros(bbox_targets.shape, dtype=np.float32)
        inds = np.where(clss > 0)[0]
        for ind in inds:
            cls = clss[ind]
            start = 4 * cls
            end = start + 4
            bbox_targets[ind, start:end] = bbox_target_data[ind, 1:]
            bbox_inside_weights[ind, start:end] = cfg.TRAIN.BBOX_INSIDE_WEIGHTS
        return bbox_targets, bbox_inside_weights
    
    def _vis_minibatch(im_blob, rois_blob, labels_blob, overlaps):
        """Visualize a mini-batch for debugging."""
        import matplotlib.pyplot as plt
        for i in xrange(rois_blob.shape[0]):
            rois = rois_blob[i, :]
            im_ind = rois[0]
            roi = rois[1:]
            im = im_blob[im_ind, :, :, :].transpose((1, 2, 0)).copy()
            im += cfg.PIXEL_MEANS
            im = im[:, :, (2, 1, 0)]
            im = im.astype(np.uint8)
            cls = labels_blob[i]
            plt.imshow(im)
            print 'class: ', cls, ' overlap: ', overlaps[i]
            plt.gca().add_patch(
                plt.Rectangle((roi[0], roi[1]), roi[2] - roi[0],
                              roi[3] - roi[1], fill=False,
                              edgecolor='r', linewidth=3)
                )
            plt.show()

lib/utils/bolb.py

# --------------------------------------------------------
    # Fast R-CNN
    # Copyright (c) 2015 Microsoft
    # Licensed under The MIT License [see LICENSE for details]
    # Written by Ross Girshick
    # --------------------------------------------------------
    
    """Blob helper functions."""
    
    import numpy as np
    import cv2
    
    def im_list_to_blob(ims):
        """Convert a list of images into a network input.
    
        Assumes images are already prepared (means subtracted, BGR order, ...).
        """
        max_shape = np.array([im.shape for im in ims]).max(axis=0)
        num_images = len(ims)
        blob = np.zeros((num_images, max_shape[0], max_shape[1], 3),
                        dtype=np.float32)
        for i in xrange(num_images):
            im = ims[i]
            blob[i, 0:im.shape[0], 0:im.shape[1], :] = im
        # Move channels (axis 3) to axis 1
        # Axis order will become: (batch elem, channel, height, width)
        channel_swap = (0, 3, 1, 2)
        blob = blob.transpose(channel_swap)
        return blob
    
    def prep_im_for_blob(im, pixel_means, target_size, max_size):
        """Mean subtract and scale an image for use in a blob."""
        im = im.astype(np.float32, copy=False)
        im -= pixel_means
        im_shape = im.shape
        im_size_min = np.min(im_shape[0:2])
        im_size_max = np.max(im_shape[0:2])
        im_scale = float(target_size) / float(im_size_min)
        # Prevent the biggest axis from being more than MAX_SIZE
        if np.round(im_scale * im_size_max) > max_size:
            im_scale = float(max_size) / float(im_size_max)
        im = cv2.resize(im, None, None, fx=im_scale, fy=im_scale,
                        interpolation=cv2.INTER_LINEAR)
    
        return im, im_scale

lib/rpn/anchor\_target\_layer.py

#coding:utf-8
    # --------------------------------------------------------
    # Faster R-CNN
    # Copyright (c) 2015 Microsoft
    # Licensed under The MIT License [see LICENSE for details]
    # Written by Ross Girshick and Sean Bell
    # --------------------------------------------------------
    
    import os
    import caffe
    import yaml
    from fast_rcnn.config import cfg
    import numpy as np
    import numpy.random as npr
    from generate_anchors import generate_anchors
    from utils.cython_bbox import bbox_overlaps
    from fast_rcnn.bbox_transform import bbox_transform
    
    DEBUG = False
    
    class AnchorTargetLayer(caffe.Layer):
        """
        Assign anchors to ground-truth targets. Produces anchor classification
        labels and bounding-box regression targets.
        """
    
        def setup(self, bottom, top):
            layer_params = yaml.load(self.param_str_)
    #设定anchor的三个尺度
            anchor_scales = layer_params.get('scales', (8, 16, 32))
    #以(8.5,8.5)为中心产生9个基准anchor
            self._anchors = generate_anchors(scales=np.array(anchor_scales))
            self._num_anchors = self._anchors.shape[0]
    #其余的anchor以feat_stride为步长上下滑动产生，config.py里feat_stride设为16，为什么是16，
    #因为不管是VGG还是ZF，conv5之后的scale是原图的1/16，这样产生的achor基本均匀分布在整个原图
            self._feat_stride = layer_params['feat_stride']
    
            if DEBUG:
                print 'anchors:'
                print self._anchors
                print 'anchor shapes:'
                print np.hstack((
                    self._anchors[:, 2::4] - self._anchors[:, 0::4],
                    self._anchors[:, 3::4] - self._anchors[:, 1::4],
                ))
                self._counts = cfg.EPS
                self._sums = np.zeros((1, 4))
                self._squared_sums = np.zeros((1, 4))
                self._fg_sum = 0
                self._bg_sum = 0
                self._count = 0
    
            # allow boxes to sit over the edge by a small amount
            self._allowed_border = layer_params.get('allowed_border', 0)
    #获得featuremap的宽高
            height, width = bottom[0].data.shape[-2:]
            if DEBUG:
                print 'AnchorTargetLayer: height', height, 'width', width
    
            A = self._num_anchors
            # labels
            top[0].reshape(1, 1, A * height, width)
            # bbox_targets
            top[1].reshape(1, A * 4, height, width)
            # bbox_inside_weights
            top[2].reshape(1, A * 4, height, width)
            # bbox_outside_weights
            top[3].reshape(1, A * 4, height, width)
    
        def forward(self, bottom, top):
            # Algorithm:
            #
            # for each (H, W) location i
            #   generate 9 anchor boxes centered on cell i
            #   apply predicted bbox deltas at cell i to each of the 9 anchors
            # filter out-of-image anchors
            # measure GT overlap
    
            assert bottom[0].data.shape[0] == 1, \
                'Only single item batches are supported'
    
            # map of shape (..., H, W)
            height, width = bottom[0].data.shape[-2:]
            # GT boxes (x1, y1, x2, y2, label)
            gt_boxes = bottom[1].data
            # im_info
            im_info = bottom[2].data[0, :]
    
            if DEBUG:
                print ''
                print 'im_size: ({}, {})'.format(im_info[0], im_info[1])
                print 'scale: {}'.format(im_info[2])
                print 'height, width: ({}, {})'.format(height, width)
                print 'rpn: gt_boxes.shape', gt_boxes.shape
                print 'rpn: gt_boxes', gt_boxes
    
            # 1. Generate proposals from bbox deltas and shifted anchors
            shift_x = np.arange(0, width) * self._feat_stride
            shift_y = np.arange(0, height) * self._feat_stride
            shift_x, shift_y = np.meshgrid(shift_x, shift_y)
            shifts = np.vstack((shift_x.ravel(), shift_y.ravel(),
                                shift_x.ravel(), shift_y.ravel())).transpose()
            # add A anchors (1, A, 4) to
            # cell K shifts (K, 1, 4) to get
            # shift anchors (K, A, 4)
            # reshape to (K*A, 4) shifted anchors
            A = self._num_anchors
            K = shifts.shape[0]
            all_anchors = (self._anchors.reshape((1, A, 4)) +
                           shifts.reshape((1, K, 4)).transpose((1, 0, 2)))
            all_anchors = all_anchors.reshape((K * A, 4))
            total_anchors = int(K * A)
    
            # only keep anchors inside the image
            inds_inside = np.where(
                (all_anchors[:, 0] >= -self._allowed_border) &
                (all_anchors[:, 1] >= -self._allowed_border) &
                (all_anchors[:, 2] < im_info[1] + self._allowed_border) &  # width
                (all_anchors[:, 3] < im_info[0] + self._allowed_border)    # height
            )[0]
    
            if DEBUG:
                print 'total_anchors', total_anchors
                print 'inds_inside', len(inds_inside)
    #裁掉大小超出图片的anchor,inds_inside是在图像内部的anchor的索引数组
            # keep only inside anchors
            anchors = all_anchors[inds_inside, :]
            if DEBUG:
                print 'anchors.shape', anchors.shape
    
            # label: 1 is positive, 0 is negative, -1 is dont care
            labels = np.empty((len(inds_inside), ), dtype=np.float32)
            labels.fill(-1)
    
            # overlaps between the anchors and the gt boxes
            # overlaps (ex, gt)
            overlaps = bbox_overlaps(
                np.ascontiguousarray(anchors, dtype=np.float),
                np.ascontiguousarray(gt_boxes, dtype=np.float))
            argmax_overlaps = overlaps.argmax(axis=1)
            max_overlaps = overlaps[np.arange(len(inds_inside)), argmax_overlaps]
            gt_argmax_overlaps = overlaps.argmax(axis=0)
            gt_max_overlaps = overlaps[gt_argmax_overlaps,
                                       np.arange(overlaps.shape[1])]
            gt_argmax_overlaps = np.where(overlaps == gt_max_overlaps)[0]
    
            if not cfg.TRAIN.RPN_CLOBBER_POSITIVES:
                # assign bg labels first so that positive labels can clobber them
                labels[max_overlaps < cfg.TRAIN.RPN_NEGATIVE_OVERLAP] = 0
    
            # fg label: for each gt, anchor with highest overlap
            labels[gt_argmax_overlaps] = 1
    
            # fg label: above threshold IOU
            labels[max_overlaps >= cfg.TRAIN.RPN_POSITIVE_OVERLAP] = 1
    
            if cfg.TRAIN.RPN_CLOBBER_POSITIVES:
                # assign bg labels last so that negative labels can clobber positives
                labels[max_overlaps < cfg.TRAIN.RPN_NEGATIVE_OVERLAP] = 0
    #采样正负anchor，如果正负样本数量不均衡，需要保持正负样本的比例基本为1：1，太悬殊
    #会使得算法漏检严重，下面的算法没有实现保持正负样本均衡
            # subsample positive labels if we have too many
            num_fg = int(cfg.TRAIN.RPN_FG_FRACTION * cfg.TRAIN.RPN_BATCHSIZE)
            fg_inds = np.where(labels == 1)[0]
            if len(fg_inds) > num_fg:
                disable_inds = npr.choice(
                    fg_inds, size=(len(fg_inds) - num_fg), replace=False)
                labels[disable_inds] = -1
    
            # subsample negative labels if we have too many
            num_bg = cfg.TRAIN.RPN_BATCHSIZE - np.sum(labels == 1)
            bg_inds = np.where(labels == 0)[0]
            if len(bg_inds) > num_bg:
                disable_inds = npr.choice(
                    bg_inds, size=(len(bg_inds) - num_bg), replace=False)
                labels[disable_inds] = -1
                #print "was %s inds, disabling %s, now %s inds" % (
                    #len(bg_inds), len(disable_inds), np.sum(labels == 0))
    
            bbox_targets = np.zeros((len(inds_inside), 4), dtype=np.float32)
            bbox_targets = _compute_targets(anchors, gt_boxes[argmax_overlaps, :])
    
            bbox_inside_weights = np.zeros((len(inds_inside), 4), dtype=np.float32)
            bbox_inside_weights[labels == 1, :] = np.array(cfg.TRAIN.RPN_BBOX_INSIDE_WEIGHTS)
    
            bbox_outside_weights = np.zeros((len(inds_inside), 4), dtype=np.float32)
            if cfg.TRAIN.RPN_POSITIVE_WEIGHT < 0:
                # uniform weighting of examples (given non-uniform sampling)
                num_examples = np.sum(labels >= 0)
                positive_weights = np.ones((1, 4)) * 1.0 / num_examples
                negative_weights = np.ones((1, 4)) * 1.0 / num_examples
            else:
                assert ((cfg.TRAIN.RPN_POSITIVE_WEIGHT > 0) &
                        (cfg.TRAIN.RPN_POSITIVE_WEIGHT < 1))
                positive_weights = (cfg.TRAIN.RPN_POSITIVE_WEIGHT /
                                    np.sum(labels == 1))
                negative_weights = ((1.0 - cfg.TRAIN.RPN_POSITIVE_WEIGHT) /
                                    np.sum(labels == 0))
            bbox_outside_weights[labels == 1, :] = positive_weights
            bbox_outside_weights[labels == 0, :] = negative_weights
    
            if DEBUG:
                self._sums += bbox_targets[labels == 1, :].sum(axis=0)
                self._squared_sums += (bbox_targets[labels == 1, :] ** 2).sum(axis=0)
                self._counts += np.sum(labels == 1)
                means = self._sums / self._counts
                stds = np.sqrt(self._squared_sums / self._counts - means ** 2)
                print 'means:'
                print means
                print 'stdevs:'
                print stds
    
            # map up to original set of anchors
            labels = _unmap(labels, total_anchors, inds_inside, fill=-1)
            bbox_targets = _unmap(bbox_targets, total_anchors, inds_inside, fill=0)
            bbox_inside_weights = _unmap(bbox_inside_weights, total_anchors, inds_inside, fill=0)
            bbox_outside_weights = _unmap(bbox_outside_weights, total_anchors, inds_inside, fill=0)
    
            if DEBUG:
                print 'rpn: max max_overlap', np.max(max_overlaps)
                print 'rpn: num_positive', np.sum(labels == 1)
                print 'rpn: num_negative', np.sum(labels == 0)
                self._fg_sum += np.sum(labels == 1)
                self._bg_sum += np.sum(labels == 0)
                self._count += 1
                print 'rpn: num_positive avg', self._fg_sum / self._count
                print 'rpn: num_negative avg', self._bg_sum / self._count
    
            # labels
            labels = labels.reshape((1, height, width, A)).transpose(0, 3, 1, 2)
            labels = labels.reshape((1, 1, A * height, width))
            top[0].reshape(*labels.shape)
            top[0].data[...] = labels
    
            # bbox_targets
            bbox_targets = bbox_targets \
                .reshape((1, height, width, A * 4)).transpose(0, 3, 1, 2)
            top[1].reshape(*bbox_targets.shape)
            top[1].data[...] = bbox_targets
    
            # bbox_inside_weights
            bbox_inside_weights = bbox_inside_weights \
                .reshape((1, height, width, A * 4)).transpose(0, 3, 1, 2)
            assert bbox_inside_weights.shape[2] == height
            assert bbox_inside_weights.shape[3] == width
            top[2].reshape(*bbox_inside_weights.shape)
            top[2].data[...] = bbox_inside_weights
    
            # bbox_outside_weights
            bbox_outside_weights = bbox_outside_weights \
                .reshape((1, height, width, A * 4)).transpose(0, 3, 1, 2)
            assert bbox_outside_weights.shape[2] == height
            assert bbox_outside_weights.shape[3] == width
            top[3].reshape(*bbox_outside_weights.shape)
            top[3].data[...] = bbox_outside_weights
    
        def backward(self, top, propagate_down, bottom):
            """This layer does not propagate gradients."""
            pass
    
        def reshape(self, bottom, top):
            """Reshaping happens during the call to forward."""
            pass
    
    
    def _unmap(data, count, inds, fill=0):
        """ Unmap a subset of item (data) back to the original set of items (of
        size count) """
        if len(data.shape) == 1:
            ret = np.empty((count, ), dtype=np.float32)
            ret.fill(fill)
            ret[inds] = data
        else:
            ret = np.empty((count, ) + data.shape[1:], dtype=np.float32)
            ret.fill(fill)
            ret[inds, :] = data
        return ret
    
    
    def _compute_targets(ex_rois, gt_rois):
        """Compute bounding-box regression targets for an image."""
    
        assert ex_rois.shape[0] == gt_rois.shape[0]
        assert ex_rois.shape[1] == 4
        assert gt_rois.shape[1] == 5
    
        return bbox_transform(ex_rois, gt_rois[:, :4]).astype(np.float32, copy=False)

用到了lib/rpn/generate\_anchors.py里的函数

rpn\_test.pt

#用RPN产生region proposal时的网络结构，这个网络只用前向计算
    name: "ZF"
    
    input: "data"
    input_shape {
      dim: 1
      dim: 3
      dim: 224
      dim: 224
    }
    
    input: "im_info"
    input_shape {
      dim: 1
      dim: 3
    }
    #前面是ZF网，特征提取用，共享
    # ------------------------ layer 1 -----------------------------
    layer {
    	name: "conv1"
    	type: "Convolution"
    	bottom: "data"
    	top: "conv1"
    	convolution_param {
    		num_output: 96
    		kernel_size: 7
    		pad: 3
    		stride: 2
    	}
    }
    layer {
    	name: "relu1"
    	type: "ReLU"
    	bottom: "conv1"
    	top: "conv1"
    }
    layer {
    	name: "norm1"
    	type: "LRN"
    	bottom: "conv1"
    	top: "norm1"
    	lrn_param {
    		local_size: 3
    		alpha: 0.00005
    		beta: 0.75
    		norm_region: WITHIN_CHANNEL
        engine: CAFFE
    	}
    }
    layer {
    	name: "pool1"
    	type: "Pooling"
    	bottom: "norm1"
    	top: "pool1"
    	pooling_param {
    		kernel_size: 3
    		stride: 2
    		pad: 1
    		pool: MAX
    	}
    }
    layer {
    	name: "conv2"
    	type: "Convolution"
    	bottom: "pool1"
    	top: "conv2"
    	convolution_param {
    		num_output: 256
    		kernel_size: 5
    		pad: 2
    		stride: 2
    	}
    }
    layer {
    	name: "relu2"
    	type: "ReLU"
    	bottom: "conv2"
    	top: "conv2"
    }
    
    layer {
    	name: "norm2"
    	type: "LRN"
    	bottom: "conv2"
    	top: "norm2"
    	lrn_param {
    		local_size: 3
    		alpha: 0.00005
    		beta: 0.75
    		norm_region: WITHIN_CHANNEL
        engine: CAFFE
    	}
    }
    layer {
    	name: "pool2"
    	type: "Pooling"
    	bottom: "norm2"
    	top: "pool2"
    	pooling_param {
    		kernel_size: 3
    		stride: 2
    		pad: 1
    		pool: MAX
    	}
    }
    layer {
    	name: "conv3"
    	type: "Convolution"
    	bottom: "pool2"
    	top: "conv3"
    	convolution_param {
    		num_output: 384
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu3"
    	type: "ReLU"
    	bottom: "conv3"
    	top: "conv3"
    }
    layer {
    	name: "conv4"
    	type: "Convolution"
    	bottom: "conv3"
    	top: "conv4"
    	convolution_param {
    		num_output: 384
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu4"
    	type: "ReLU"
    	bottom: "conv4"
    	top: "conv4"
    }
    layer {
    	name: "conv5"
    	type: "Convolution"
    	bottom: "conv4"
    	top: "conv5"
    	convolution_param {
    		num_output: 256#经过最后一层，产生256个特征图
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu5"
    	type: "ReLU"
    	bottom: "conv5"
    	top: "conv5"
    }
    
    #-----------------------layer +-------------------------
    #RPN在conv5上滑动窗口，256*3*3*256卷积核，预测每个位置9个anchor是否属于前景，
    #如果属于前景，box的修正位置
    layer {
      name: "rpn_conv1"
      type: "Convolution"
      bottom: "conv5"
      top: "rpn_conv1"
      convolution_param {
        num_output: 256
        kernel_size: 3 pad: 1 stride: 1
      }
    }
    layer {
      name: "rpn_relu1"
      type: "ReLU"
      bottom: "rpn_conv1"
      top: "rpn_conv1"
    }
    layer {
      name: "rpn_cls_score"
      type: "Convolution"
      bottom: "rpn_conv1"
      top: "rpn_cls_score"
      convolution_param {
        num_output: 18   # 2(bg/fg) * 9(anchors)#输出预测每个位置9个anchor，属于bg或fg
        kernel_size: 1 pad: 0 stride: 1
      }
    }
    layer {
      name: "rpn_bbox_pred"
      type: "Convolution"
      bottom: "rpn_conv1"
      top: "rpn_bbox_pred"
      convolution_param {
        num_output: 36   # 4 * 9(anchors)#输出预测9个anchor的修正坐标
        kernel_size: 1 pad: 0 stride: 1
      }
    }
    layer {
       bottom: "rpn_cls_score"
       top: "rpn_cls_score_reshape"
       name: "rpn_cls_score_reshape"
       type: "Reshape"
       reshape_param { shape { dim: 0 dim: 2 dim: -1 dim: 0 } }
    }
    
    #-----------------------output------------------------
    layer {
      name: "rpn_cls_prob"
      type: "Softmax"
      bottom: "rpn_cls_score_reshape"
      top: "rpn_cls_prob"
    }
    layer {
      name: 'rpn_cls_prob_reshape'
      type: 'Reshape'
      bottom: 'rpn_cls_prob'
      top: 'rpn_cls_prob_reshape'
      reshape_param { shape { dim: 0 dim: 18 dim: -1 dim: 0 } }
    }
    layer {
      name: 'proposal'
      type: 'Python'
      bottom: 'rpn_cls_prob_reshape'
      bottom: 'rpn_bbox_pred'
      bottom: 'im_info'
      top: 'rois'
      top: 'scores'
      python_param {
        module: 'rpn.proposal_layer'#对应lib/rpn/proposal_layer.py
        layer: 'ProposalLayer'
        param_str: "'feat_stride': 16"
      }
    }

lib/rpn/proposal\_layer.py

这一层用来由RPN产生region proposal

fast\_rcnn\_train.pt

#stage 1训练fast rcnn网络，输入是rpn提取的roi以及gt box
    name: "ZF"
    layer {
      name: 'data'
      type: 'Python'
      top: 'data'
      top: 'rois'
      top: 'labels'
      top: 'bbox_targets'
      top: 'bbox_inside_weights'
      top: 'bbox_outside_weights'
      python_param {
        module: 'roi_data_layer.layer'#对应lib/roi_data_layer/layer.py
    #为训练fast rcnn时为网络输入roi，此时为roi是region proposal
        layer: 'RoIDataLayer'
        param_str: "'num_classes': 21"
      }
    }
    
    #ZF网，特征提取用，共享
    #========= conv1-conv5 ============
    
    layer {
    	name: "conv1"
    	type: "Convolution"
    	bottom: "data"
    	top: "conv1"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 96
    		kernel_size: 7
    		pad: 3
    		stride: 2
    	}
    }
    layer {
    	name: "relu1"
    	type: "ReLU"
    	bottom: "conv1"
    	top: "conv1"
    }
    layer {
    	name: "norm1"
    	type: "LRN"
    	bottom: "conv1"
    	top: "norm1"
    	lrn_param {
    		local_size: 3
    		alpha: 0.00005
    		beta: 0.75
    		norm_region: WITHIN_CHANNEL
        engine: CAFFE
    	}
    }
    layer {
    	name: "pool1"
    	type: "Pooling"
    	bottom: "norm1"
    	top: "pool1"
    	pooling_param {
    		kernel_size: 3
    		stride: 2
    		pad: 1
    		pool: MAX
    	}
    }
    layer {
    	name: "conv2"
    	type: "Convolution"
    	bottom: "pool1"
    	top: "conv2"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 256
    		kernel_size: 5
    		pad: 2
    		stride: 2
    	}
    }
    layer {
    	name: "relu2"
    	type: "ReLU"
    	bottom: "conv2"
    	top: "conv2"
    }
    layer {
    	name: "norm2"
    	type: "LRN"
    	bottom: "conv2"
    	top: "norm2"
    	lrn_param {
    		local_size: 3
    		alpha: 0.00005
    		beta: 0.75
    		norm_region: WITHIN_CHANNEL
        engine: CAFFE
    	}
    }
    layer {
    	name: "pool2"
    	type: "Pooling"
    	bottom: "norm2"
    	top: "pool2"
    	pooling_param {
    		kernel_size: 3
    		stride: 2
    		pad: 1
    		pool: MAX
    	}
    }
    layer {
    	name: "conv3"
    	type: "Convolution"
    	bottom: "pool2"
    	top: "conv3"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 384
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu3"
    	type: "ReLU"
    	bottom: "conv3"
    	top: "conv3"
    }
    layer {
    	name: "conv4"
    	type: "Convolution"
    	bottom: "conv3"
    	top: "conv4"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 384
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu4"
    	type: "ReLU"
    	bottom: "conv4"
    	top: "conv4"
    }
    layer {
    	name: "conv5"
    	type: "Convolution"
    	bottom: "conv4"
    	top: "conv5"
    	param { lr_mult: 1.0 }
    	param { lr_mult: 2.0 }
    	convolution_param {
    		num_output: 256
    		kernel_size: 3
    		pad: 1
    		stride: 1
    	}
    }
    layer {
    	name: "relu5"
    	type: "ReLU"
    	bottom: "conv5"
    	top: "conv5"
    }
    
    #========= RCNN ============
    
    layer {
      name: "roi_pool_conv5"
      type: "ROIPooling"#这个层在caffe-fast-rcnn里实现
      bottom: "conv5"
      bottom: "rois"
      top: "roi_pool_conv5"
      roi_pooling_param {#每个roi做max pooling后的大小为6*6
        pooled_w: 6
        pooled_h: 6
        spatial_scale: 0.0625 # 1/16
      }
    }
    layer {
      name: "fc6"
      type: "InnerProduct"
      bottom: "roi_pool_conv5"
      top: "fc6"
      param { lr_mult: 1.0 }
      param { lr_mult: 2.0 }
      inner_product_param {
        num_output: 4096
      }
    }
    layer {
      name: "relu6"
      type: "ReLU"
      bottom: "fc6"
      top: "fc6"
    }
    layer {
      name: "drop6"
      type: "Dropout"
      bottom: "fc6"
      top: "fc6"
      dropout_param {
        dropout_ratio: 0.5
        scale_train: false
      }
    }
    layer {
      name: "fc7"
      type: "InnerProduct"
      bottom: "fc6"
      top: "fc7"
      param { lr_mult: 1.0 }
      param { lr_mult: 2.0 }
      inner_product_param {
        num_output: 4096
      }
    }
    layer {
      name: "relu7"
      type: "ReLU"
      bottom: "fc7"
      top: "fc7"
    }
    layer {
      name: "drop7"
      type: "Dropout"
      bottom: "fc7"
      top: "fc7"
      dropout_param {
        dropout_ratio: 0.5
        scale_train: false
      }
    }
    layer {
      name: "cls_score"
      type: "InnerProduct"
      bottom: "fc7"
      top: "cls_score"
      param { lr_mult: 1.0 }
      param { lr_mult: 2.0 }
      inner_product_param {
        num_output: 21
        weight_filler {
          type: "gaussian"
          std: 0.01
        }
        bias_filler {
          type: "constant"
          value: 0
        }
      }
    }
    layer {
      name: "bbox_pred"
      type: "InnerProduct"
      bottom: "fc7"
      top: "bbox_pred"
      param { lr_mult: 1.0 }
      param { lr_mult: 2.0 }
      inner_product_param {
        num_output: 84
        weight_filler {
          type: "gaussian"
          std: 0.001
        }
        bias_filler {
          type: "constant"
          value: 0
        }
      }
    }
    layer {
      name: "loss_cls"
      type: "SoftmaxWithLoss"
      bottom: "cls_score"
      bottom: "labels"
      propagate_down: 1
      propagate_down: 0
      top: "cls_loss"
      loss_weight: 1
      loss_param {
        ignore_label: -1
        normalize: true
      }
    }
    layer {
      name: "loss_bbox"
      type: "SmoothL1Loss"
      bottom: "bbox_pred"
      bottom: "bbox_targets"
      bottom: "bbox_inside_weights"
      bottom: "bbox_outside_weights"
      top: "bbox_loss"
      loss_weight: 1
    }
    
    #========= RPN ============
    # Dummy layers so that initial parameters are saved into the output net
    
    layer {
      name: "rpn_conv1"
      type: "Convolution"
      bottom: "conv5"
      top: "rpn_conv1"
      param { lr_mult: 0 decay_mult: 0 }
      param { lr_mult: 0 decay_mult: 0 }
      convolution_param {
        num_output: 256
        kernel_size: 3 pad: 1 stride: 1
        weight_filler { type: "gaussian" std: 0.01 }
        bias_filler { type: "constant" value: 0 }
      }
    }
    layer {
      name: "rpn_relu1"
      type: "ReLU"
      bottom: "rpn_conv1"
      top: "rpn_conv1"
    }
    layer {
      name: "rpn_cls_score"
      type: "Convolution"
      bottom: "rpn_conv1"
      top: "rpn_cls_score"
      param { lr_mult: 0 decay_mult: 0 }
      param { lr_mult: 0 decay_mult: 0 }
      convolution_param {
        num_output: 18   # 2(bg/fg) * 9(anchors)
        kernel_size: 1 pad: 0 stride: 1
        weight_filler { type: "gaussian" std: 0.01 }
        bias_filler { type: "constant" value: 0 }
      }
    }
    layer {
      name: "rpn_bbox_pred"
      type: "Convolution"
      bottom: "rpn_conv1"
      top: "rpn_bbox_pred"
      param { lr_mult: 0 decay_mult: 0 }
      param { lr_mult: 0 decay_mult: 0 }
      convolution_param {
        num_output: 36   # 4 * 9(anchors)
        kernel_size: 1 pad: 0 stride: 1
        weight_filler { type: "gaussian" std: 0.01 }
        bias_filler { type: "constant" value: 0 }
      }
    }
    layer {
      name: "silence_rpn_cls_score"
      type: "Silence"
      bottom: "rpn_cls_score"
    }
    layer {
      name: "silence_rpn_bbox_pred"
      type: "Silence"
      bottom: "rpn_bbox_pred"
    }

其中roi pooling layer在 caffe/src/layers/roi\_pooling\_layer.cpp里实现

// ------------------------------------------------------------------
    // Fast R-CNN
    // Copyright (c) 2015 Microsoft
    // Licensed under The MIT License [see fast-rcnn/LICENSE for details]
    // Written by Ross Girshick
    // ------------------------------------------------------------------
    
    #include <cfloat>
    
    #include "caffe/fast_rcnn_layers.hpp"
    
    using std::max;
    using std::min;
    using std::floor;
    using std::ceil;
    
    namespace caffe {
    
    template <typename Dtype>
    void ROIPoolingLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
          const vector<Blob<Dtype>*>& top) {
      ROIPoolingParameter roi_pool_param = this->layer_param_.roi_pooling_param();
      CHECK_GT(roi_pool_param.pooled_h(), 0)
          << "pooled_h must be > 0";
      CHECK_GT(roi_pool_param.pooled_w(), 0)
          << "pooled_w must be > 0";
      pooled_height_ = roi_pool_param.pooled_h();
      pooled_width_ = roi_pool_param.pooled_w();
      spatial_scale_ = roi_pool_param.spatial_scale();
      LOG(INFO) << "Spatial scale: " << spatial_scale_;
    }
    
    template <typename Dtype>
    void ROIPoolingLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
          const vector<Blob<Dtype>*>& top) {
      channels_ = bottom[0]->channels();
      height_ = bottom[0]->height();
      width_ = bottom[0]->width();
      top[0]->Reshape(bottom[1]->num(), channels_, pooled_height_,
          pooled_width_);
      max_idx_.Reshape(bottom[1]->num(), channels_, pooled_height_,
          pooled_width_);
    }
    
    template <typename Dtype>
    void ROIPoolingLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
          const vector<Blob<Dtype>*>& top) {
      const Dtype* bottom_data = bottom[0]->cpu_data();
      const Dtype* bottom_rois = bottom[1]->cpu_data();
      // Number of ROIs
      int num_rois = bottom[1]->num();
      int batch_size = bottom[0]->num();
      int top_count = top[0]->count();
      Dtype* top_data = top[0]->mutable_cpu_data();
      caffe_set(top_count, Dtype(-FLT_MAX), top_data);
      int* argmax_data = max_idx_.mutable_cpu_data();
      caffe_set(top_count, -1, argmax_data);
    
      // For each ROI R = [batch_index x1 y1 x2 y2]: max pool over R
      for (int n = 0; n < num_rois; ++n) {
        int roi_batch_ind = bottom_rois[0];
        int roi_start_w = round(bottom_rois[1] * spatial_scale_);
        int roi_start_h = round(bottom_rois[2] * spatial_scale_);
        int roi_end_w = round(bottom_rois[3] * spatial_scale_);
        int roi_end_h = round(bottom_rois[4] * spatial_scale_);
        CHECK_GE(roi_batch_ind, 0);
        CHECK_LT(roi_batch_ind, batch_size);
    
        int roi_height = max(roi_end_h - roi_start_h + 1, 1);
        int roi_width = max(roi_end_w - roi_start_w + 1, 1);
        const Dtype bin_size_h = static_cast<Dtype>(roi_height)
                                 / static_cast<Dtype>(pooled_height_);
        const Dtype bin_size_w = static_cast<Dtype>(roi_width)
                                 / static_cast<Dtype>(pooled_width_);
    
        const Dtype* batch_data = bottom_data + bottom[0]->offset(roi_batch_ind);
    
        for (int c = 0; c < channels_; ++c) {
          for (int ph = 0; ph < pooled_height_; ++ph) {
            for (int pw = 0; pw < pooled_width_; ++pw) {
              // Compute pooling region for this output unit:
              //  start (included) = floor(ph * roi_height / pooled_height_)
              //  end (excluded) = ceil((ph + 1) * roi_height / pooled_height_)
              int hstart = static_cast<int>(floor(static_cast<Dtype>(ph)
                                                  * bin_size_h));
              int wstart = static_cast<int>(floor(static_cast<Dtype>(pw)
                                                  * bin_size_w));
              int hend = static_cast<int>(ceil(static_cast<Dtype>(ph + 1)
                                               * bin_size_h));
              int wend = static_cast<int>(ceil(static_cast<Dtype>(pw + 1)
                                               * bin_size_w));
    
              hstart = min(max(hstart + roi_start_h, 0), height_);
              hend = min(max(hend + roi_start_h, 0), height_);
              wstart = min(max(wstart + roi_start_w, 0), width_);
              wend = min(max(wend + roi_start_w, 0), width_);
    
              bool is_empty = (hend <= hstart) || (wend <= wstart);
    
              const int pool_index = ph * pooled_width_ + pw;
              if (is_empty) {
                top_data[pool_index] = 0;
                argmax_data[pool_index] = -1;
              }
    
              for (int h = hstart; h < hend; ++h) {
                for (int w = wstart; w < wend; ++w) {
                  const int index = h * width_ + w;
                  if (batch_data[index] > top_data[pool_index]) {
                    top_data[pool_index] = batch_data[index];
                    argmax_data[pool_index] = index;
                  }
                }
              }
            }
          }
          // Increment all data pointers by one channel
          batch_data += bottom[0]->offset(0, 1);
          top_data += top[0]->offset(0, 1);
          argmax_data += max_idx_.offset(0, 1);
        }
        // Increment ROI data pointer
        bottom_rois += bottom[1]->offset(1);
      }
    }
    
    template <typename Dtype>
    void ROIPoolingLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
          const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom) {
      NOT_IMPLEMENTED;
    }
    
    
    #ifdef CPU_ONLY
    STUB_GPU(ROIPoolingLayer);
    #endif
    
    INSTANTIATE_CLASS(ROIPoolingLayer);
    REGISTER_LAYER_CLASS(ROIPooling);
    
    }  // namespace caffe

大致结构看明白了来看具体训练流程

首先看tools/train\_faster\_rcnn\_alt\_opt.py

#coding:utf-8
    #!/usr/bin/env python
    
    # --------------------------------------------------------
    # Faster R-CNN
    # Copyright (c) 2015 Microsoft
    # Licensed under The MIT License [see LICENSE for details]
    # Written by Ross Girshick
    # --------------------------------------------------------
    
    """Train a Faster R-CNN network using alternating optimization.
    This tool implements the alternating optimization algorithm described in our
    NIPS 2015 paper ("Faster R-CNN: Towards Real-time Object Detection with Region
    Proposal Networks." Shaoqing Ren, Kaiming He, Ross Girshick, Jian Sun.)
    """
    
    import _init_paths
    from fast_rcnn.train import get_training_roidb, train_net
    from fast_rcnn.config import cfg, cfg_from_file, cfg_from_list, get_output_dir
    from datasets.factory import get_imdb
    from rpn.generate import imdb_proposals
    import argparse
    import pprint
    import numpy as np
    import sys, os
    import multiprocessing as mp
    import cPickle
    import shutil
    
    def parse_args():
        """
        Parse input arguments
        """
        parser = argparse.ArgumentParser(description='Train a Faster R-CNN network')
        #训练时设置使用哪个GPU
        parser.add_argument('--gpu', dest='gpu_id',
                            help='GPU device id to use [0]',
                            default=0, type=int)
        #设置训练时使用哪种网络模型
        parser.add_argument('--net_name', dest='net_name',
                            help='network name (e.g., "ZF")',
                            default=None, type=str)
        #指定预训练的模型来初始化网络
        parser.add_argument('--weights', dest='pretrained_model',
                            help='initialize with pretrained model weights',
                            default=None, type=str)
        #加载配置文件
        parser.add_argument('--cfg', dest='cfg_file',
                            help='optional config file',
                            default=None, type=str)
        #训练使用的数据集
        parser.add_argument('--imdb', dest='imdb_name',
                            help='dataset to train on',
                            default='voc_2007_trainval', type=str)
        parser.add_argument('--set', dest='set_cfgs',
                            help='set config keys', default=None,
                            nargs=argparse.REMAINDER)
    
        if len(sys.argv) == 1:
            parser.print_help()
            sys.exit(1)
    
        args = parser.parse_args()
        return args
    
    def get_roidb(imdb_name, rpn_file=None):
    #得到图像集（image database）的名字，如pascalvoc——2007——trainval
        imdb = get_imdb(imdb_name)
        print 'Loaded dataset `{:s}` for training'.format(imdb.name)
    #设置网络得到proposal的方法，有selective search和RPN、gt，selective search已弃用
        imdb.set_proposal_method(cfg.TRAIN.PROPOSAL_METHOD)
        print 'Set proposal method: {:s}'.format(cfg.TRAIN.PROPOSAL_METHOD)
    #判断之前是否已经有RPN网络提取得到的region proposal文件
        if rpn_file is not None:
            imdb.config['rpn_file'] = rpn_file
        roidb = get_training_roidb(imdb)
        return roidb, imdb
    
    def get_solvers(net_name):
        # Faster R-CNN Alternating Optimization
        n = 'faster_rcnn_alt_opt'
        # Solver for each training stage
        solvers = [[net_name, n, 'stage1_rpn_solver60k80k.pt'],
                   [net_name, n, 'stage1_fast_rcnn_solver30k40k.pt'],
                   [net_name, n, 'stage2_rpn_solver60k80k.pt'],
                   [net_name, n, 'stage2_fast_rcnn_solver30k40k.pt']]
        solvers = [os.path.join(cfg.MODELS_DIR, *s) for s in solvers]
        # Iterations for each training stage
    #每一轮训练的最大迭代次数，建议测试时都设置为100
        max_iters = [80000, 40000, 80000, 40000]
        # max_iters = [100, 100, 100, 100]
        # Test prototxt for the RPN
        rpn_test_prototxt = os.path.join(
            cfg.MODELS_DIR, net_name, n, 'rpn_test.pt')
        return solvers, max_iters, rpn_test_prototxt
    
    # ------------------------------------------------------------------------------
    # Pycaffe doesn't reliably free GPU memory when instantiated nets are discarded
    # (e.g. "del net" in Python code). To work around this issue, each training
    # stage is executed in a separate process using multiprocessing.Process.
    # ------------------------------------------------------------------------------
    
    def _init_caffe(cfg):
        """Initialize pycaffe in a training process.
        """
    
        import caffe
        # fix the random seeds (numpy and caffe) for reproducibility
        np.random.seed(cfg.RNG_SEED)
        caffe.set_random_seed(cfg.RNG_SEED)
        # set up caffe
        caffe.set_mode_gpu()
        caffe.set_device(cfg.GPU_ID)
    
    #训练RPN
    def train_rpn(queue=None, imdb_name=None, init_model=None, solver=None,
                  max_iters=None, cfg=None):
        """Train a Region Proposal Network in a separate training process.
        """
    
        # Not using any proposals, just ground-truth boxes
        cfg.TRAIN.HAS_RPN = True
        cfg.TRAIN.BBOX_REG = False  # applies only to Fast R-CNN bbox regression
    #训练RPN时使用ground-truth
        cfg.TRAIN.PROPOSAL_METHOD = 'gt'
    #每次训练RPN只用一张图片
        cfg.TRAIN.IMS_PER_BATCH = 1
        print 'Init model: {}'.format(init_model)
        print('Using config:')
        pprint.pprint(cfg)
    
        import caffe
        _init_caffe(cfg)
    
        roidb, imdb = get_roidb(imdb_name)
        print 'roidb len: {}'.format(len(roidb))
        output_dir = get_output_dir(imdb)
        print 'Output will be saved to `{:s}`'.format(output_dir)
    #开始训练RPN网络
        model_paths = train_net(solver, roidb, output_dir,
                                pretrained_model=init_model,
                                max_iters=max_iters)
    #只保留最后得到的网络模型
        # Cleanup all but the final model
        for i in model_paths[:-1]:
            os.remove(i)
        rpn_model_path = model_paths[-1]
        # Send final model path through the multiprocessing queue
        queue.put({'model_path': rpn_model_path})
    
    
    #用训练完的RPN产生region proposal并存到磁盘上
    def rpn_generate(queue=None, imdb_name=None, rpn_model_path=None, cfg=None,
                     rpn_test_prototxt=None):
        """Use a trained RPN to generate proposals.
        """
    
        cfg.TEST.RPN_PRE_NMS_TOP_N = -1     # no pre NMS filtering
        cfg.TEST.RPN_POST_NMS_TOP_N = 2000  # limit top boxes after NMS
        print 'RPN model: {}'.format(rpn_model_path)
        print('Using config:')
        pprint.pprint(cfg)
    
        import caffe
        _init_caffe(cfg)
    
        # NOTE: the matlab implementation computes proposals on flipped images, too.
        # We compute them on the image once and then flip the already computed
        # proposals. This might cause a minor loss in mAP (less proposal jittering).
        imdb = get_imdb(imdb_name)
        print 'Loaded dataset `{:s}` for proposal generation'.format(imdb.name)
    
        # Load RPN and configure output directory
        rpn_net = caffe.Net(rpn_test_prototxt, rpn_model_path, caffe.TEST)
        output_dir = get_output_dir(imdb)
        print 'Output will be saved to `{:s}`'.format(output_dir)
        # Generate proposals on the imdb
        rpn_proposals = imdb_proposals(rpn_net, imdb)
        # Write proposals to disk and send the proposal file path through the
        # multiprocessing queue
        rpn_net_name = os.path.splitext(os.path.basename(rpn_model_path))[0]
        rpn_proposals_path = os.path.join(
            output_dir, rpn_net_name + '_proposals.pkl')
        with open(rpn_proposals_path, 'wb') as f:
            cPickle.dump(rpn_proposals, f, cPickle.HIGHEST_PROTOCOL)
        print 'Wrote RPN proposals to {}'.format(rpn_proposals_path)
        queue.put({'proposal_path': rpn_proposals_path})
    
    #训练fast-rcnn
    def train_fast_rcnn(queue=None, imdb_name=None, init_model=None, solver=None,
                        max_iters=None, cfg=None, rpn_file=None):
        """Train a Fast R-CNN using proposals generated by an RPN.
        """
    #conv5后面现在接的是fast-rcnn
        cfg.TRAIN.HAS_RPN = False           # not generating prosals on-the-fly
    #roidb由刚刚训练完的RPN产生
        cfg.TRAIN.PROPOSAL_METHOD = 'rpn'   # use pre-computed RPN proposals instead
    #每次训练fast-rcnn使用两张图片
        cfg.TRAIN.IMS_PER_BATCH = 2
        print 'Init model: {}'.format(init_model)
        print 'RPN proposals: {}'.format(rpn_file)
        print('Using config:')
        pprint.pprint(cfg)
    
        import caffe
        _init_caffe(cfg)
    
        roidb, imdb = get_roidb(imdb_name, rpn_file=rpn_file)
        output_dir = get_output_dir(imdb)
        print 'Output will be saved to `{:s}`'.format(output_dir)
        # Train Fast R-CNN
        model_paths = train_net(solver, roidb, output_dir,
                                pretrained_model=init_model,
                                max_iters=max_iters)
        # Cleanup all but the final model
        for i in model_paths[:-1]:
            os.remove(i)
        fast_rcnn_model_path = model_paths[-1]
        # Send Fast R-CNN model path over the multiprocessing queue
        queue.put({'model_path': fast_rcnn_model_path})
    
    if __name__ == '__main__':
        args = parse_args()
    
        print('Called with args:')
        print(args)
    
        if args.cfg_file is not None:
            cfg_from_file(args.cfg_file)
        if args.set_cfgs is not None:
            cfg_from_list(args.set_cfgs)
        cfg.GPU_ID = args.gpu_id
    
        # --------------------------------------------------------------------------
        # Pycaffe doesn't reliably free GPU memory when instantiated nets are
        # discarded (e.g. "del net" in Python code). To work around this issue, each
        # training stage is executed in a separate process using
        # multiprocessing.Process.
        # --------------------------------------------------------------------------
    
        # queue for communicated results between processes
        mp_queue = mp.Queue()
        # solves, iters, etc. for each training stage
        solvers, max_iters, rpn_test_prototxt = get_solvers(args.net_name)
    
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
        print 'Stage 1 RPN, init from ImageNet model'
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
    
        cfg.TRAIN.SNAPSHOT_INFIX = 'stage1'
        mp_kwargs = dict(
                queue=mp_queue,
                imdb_name=args.imdb_name,
                init_model=args.pretrained_model,
                solver=solvers[0],
                max_iters=max_iters[0],
                cfg=cfg)
        p = mp.Process(target=train_rpn, kwargs=mp_kwargs)
        p.start()
        rpn_stage1_out = mp_queue.get()
        p.join()
    
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
        print 'Stage 1 RPN, generate proposals'
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
    
        mp_kwargs = dict(
                queue=mp_queue,
                imdb_name=args.imdb_name,
                rpn_model_path=str(rpn_stage1_out['model_path']),
                cfg=cfg,
                rpn_test_prototxt=rpn_test_prototxt)
        p = mp.Process(target=rpn_generate, kwargs=mp_kwargs)
        p.start()
        rpn_stage1_out['proposal_path'] = mp_queue.get()['proposal_path']
        p.join()
    
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
        print 'Stage 1 Fast R-CNN using RPN proposals, init from ImageNet model'
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
    
        cfg.TRAIN.SNAPSHOT_INFIX = 'stage1'
        mp_kwargs = dict(
                queue=mp_queue,
                imdb_name=args.imdb_name,
                init_model=args.pretrained_model,
                solver=solvers[1],
                max_iters=max_iters[1],
                cfg=cfg,
                rpn_file=rpn_stage1_out['proposal_path'])
        p = mp.Process(target=train_fast_rcnn, kwargs=mp_kwargs)
        p.start()
        fast_rcnn_stage1_out = mp_queue.get()
        p.join()
    
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
        print 'Stage 2 RPN, init from stage 1 Fast R-CNN model'
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
    
        cfg.TRAIN.SNAPSHOT_INFIX = 'stage2'
        mp_kwargs = dict(
                queue=mp_queue,
                imdb_name=args.imdb_name,
                init_model=str(fast_rcnn_stage1_out['model_path']),
                solver=solvers[2],
                max_iters=max_iters[2],
                cfg=cfg)
        p = mp.Process(target=train_rpn, kwargs=mp_kwargs)
        p.start()
        rpn_stage2_out = mp_queue.get()
        p.join()
    
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
        print 'Stage 2 RPN, generate proposals'
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
    
        mp_kwargs = dict(
                queue=mp_queue,
                imdb_name=args.imdb_name,
                rpn_model_path=str(rpn_stage2_out['model_path']),
                cfg=cfg,
                rpn_test_prototxt=rpn_test_prototxt)
        p = mp.Process(target=rpn_generate, kwargs=mp_kwargs)
        p.start()
        rpn_stage2_out['proposal_path'] = mp_queue.get()['proposal_path']
        p.join()
    
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
        print 'Stage 2 Fast R-CNN, init from stage 2 RPN R-CNN model'
        print '~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'
    
        cfg.TRAIN.SNAPSHOT_INFIX = 'stage2'
        mp_kwargs = dict(
                queue=mp_queue,
                imdb_name=args.imdb_name,
                init_model=str(rpn_stage2_out['model_path']),
                solver=solvers[3],
                max_iters=max_iters[3],
                cfg=cfg,
                rpn_file=rpn_stage2_out['proposal_path'])
        p = mp.Process(target=train_fast_rcnn, kwargs=mp_kwargs)
        p.start()
        fast_rcnn_stage2_out = mp_queue.get()
        p.join()
    
        # Create final model (just a copy of the last stage)
        final_path = os.path.join(
                os.path.dirname(fast_rcnn_stage2_out['model_path']),
                args.net_name + '_faster_rcnn_final.caffemodel')
        print 'cp {} -> {}'.format(
                fast_rcnn_stage2_out['model_path'], final_path)
        shutil.copy(fast_rcnn_stage2_out['model_path'], final_path)
        print 'Final model: {}'.format(final_path)

lib/rpn/generate.py利用rpn网络前向计算产生proposal

lib/fast\_rcnn/train.py

lib/roi\_data\_layer/roidb.py

roidb是一个重要的数据结构，roidb是一个列表，里面的每个元素是字典，对应一张图片的所有roi信息  
\{'image':imageindex,'width':w,'height':h,'gt\_overlaps':二维array,每张图片上所有roi与各个类别的gt的overlap，'max\_classes':max\_cls,每个roi属于那一类别的大，'max\_overlaps'：每个roi与gt最大重叠率的大小\}

lib/datasets/imdb.py

#coding:utf-8
    # --------------------------------------------------------
    # Fast R-CNN
    # Copyright (c) 2015 Microsoft
    # Licensed under The MIT License [see LICENSE for details]
    # Written by Ross Girshick
    # --------------------------------------------------------
    
    import os
    import os.path as osp
    import PIL
    from utils.cython_bbox import bbox_overlaps
    import numpy as np
    import scipy.sparse
    from fast_rcnn.config import cfg
    
    class imdb(object):
        """Image database."""
    
        def __init__(self, name):
    #imdb的一些属性   
            self._name = name
            self._num_classes = 0
            self._classes = []
            self._image_index = []
            self._obj_proposer = 'selective_search'
            self._roidb = None
            self._roidb_handler = self.default_roidb
            # Use this dict for storing dataset specific config options
            self.config = {}
    
        @property
        def name(self):
            return self._name
    
        @property
        def num_classes(self):
            return len(self._classes)
    
        @property
        def classes(self):
            return self._classes
    
        @property
        def image_index(self):
            return self._image_index
    
        @property
        def roidb_handler(self):
            return self._roidb_handler
    
        @roidb_handler.setter
        def roidb_handler(self, val):
            self._roidb_handler = val
    
        def set_proposal_method(self, method):
            method = eval('self.' + method + '_roidb')
            self.roidb_handler = method
    
        @property
        def roidb(self):
            # A roidb is a list of dictionaries, each with the following keys:
            #   boxes
            #   gt_overlaps
            #   gt_classes
            #   flipped
            if self._roidb is not None:
                return self._roidb
            self._roidb = self.roidb_handler()
            return self._roidb
    
        @property
        def cache_path(self):
            cache_path = osp.abspath(osp.join(cfg.DATA_DIR, 'cache'))
            if not os.path.exists(cache_path):
                os.makedirs(cache_path)
            return cache_path
    
        @property
        def num_images(self):
          return len(self.image_index)
    
        def image_path_at(self, i):
            raise NotImplementedError
    
        def default_roidb(self):
            raise NotImplementedError
    
        def evaluate_detections(self, all_boxes, output_dir=None):
            """
            all_boxes is a list of length number-of-classes.
            Each list element is a list of length number-of-images.
            Each of those list elements is either an empty list []
            or a numpy array of detection.
    
            all_boxes[class][image] = [] or np.array of shape #dets x 5
            """
            raise NotImplementedError
    
        def _get_widths(self):
          return [PIL.Image.open(self.image_path_at(i)).size[0]
                  for i in xrange(self.num_images)]
    #对所有原始图像进行水平翻转
        def append_flipped_images(self):
            num_images = self.num_images
    #得到所有图像的宽度存入list
            widths = self._get_widths()
            for i in xrange(num_images):
    #复制每张图中所有的box坐标，这个boxes是一个列表，类似[(x1min,y1min,x1max,y1max),]
                boxes = self.roidb[i]['boxes'].copy()
                oldx1 = boxes[:, 0].copy()
                oldx2 = boxes[:, 2].copy()
    #水平翻转只用对横坐标做变换，容易得到x'=width-x
                boxes[:, 0] = widths[i] - oldx2 - 1
                boxes[:, 2] = widths[i] - oldx1 - 1
                assert (boxes[:, 2] >= boxes[:, 0]).all()
    #entry是一个字典，存了boxes坐标，和ground-truth的重叠率，类别，是否由水平翻转得到，
    #重叠率和类别不会变，直接复制i
                entry = {'boxes' : boxes,
                         'gt_overlaps' : self.roidb[i]['gt_overlaps'],
                         'gt_classes' : self.roidb[i]['gt_classes'],
                         'flipped' : True}
    #把水平翻转得到的数据加入roidb
                self.roidb.append(entry)
    #数量变为原来的2倍
            self._image_index = self._image_index * 2
    
        def evaluate_recall(self, candidate_boxes=None, thresholds=None,
                            area='all', limit=None):
            """Evaluate detection proposal recall metrics.
    
            Returns:
                results: dictionary of results with keys
                    'ar': average recall
                    'recalls': vector recalls at each IoU overlap threshold
                    'thresholds': vector of IoU overlap thresholds
                    'gt_overlaps': vector of all ground-truth overlaps
            """
            # Record max overlap value for each gt box
            # Return vector of overlap values
            areas = { 'all': 0, 'small': 1, 'medium': 2, 'large': 3,
                      '96-128': 4, '128-256': 5, '256-512': 6, '512-inf': 7}
            area_ranges = [ [0**2, 1e5**2],    # all
                            [0**2, 32**2],     # small
                            [32**2, 96**2],    # medium
                            [96**2, 1e5**2],   # large
                            [96**2, 128**2],   # 96-128
                            [128**2, 256**2],  # 128-256
                            [256**2, 512**2],  # 256-512
                            [512**2, 1e5**2],  # 512-inf
                          ]
            assert areas.has_key(area), 'unknown area range: {}'.format(area)
            area_range = area_ranges[areas[area]]
            gt_overlaps = np.zeros(0)
            num_pos = 0
            for i in xrange(self.num_images):
                # Checking for max_overlaps == 1 avoids including crowd annotations
                # (...pretty hacking :/)
                max_gt_overlaps = self.roidb[i]['gt_overlaps'].toarray().max(axis=1)
                gt_inds = np.where((self.roidb[i]['gt_classes'] > 0) &
                                   (max_gt_overlaps == 1))[0]
                gt_boxes = self.roidb[i]['boxes'][gt_inds, :]
                gt_areas = self.roidb[i]['seg_areas'][gt_inds]
                valid_gt_inds = np.where((gt_areas >= area_range[0]) &
                                         (gt_areas <= area_range[1]))[0]
                gt_boxes = gt_boxes[valid_gt_inds, :]
                num_pos += len(valid_gt_inds)
    
                if candidate_boxes is None:
                    # If candidate_boxes is not supplied, the default is to use the
                    # non-ground-truth boxes from this roidb
                    non_gt_inds = np.where(self.roidb[i]['gt_classes'] == 0)[0]
                    boxes = self.roidb[i]['boxes'][non_gt_inds, :]
                else:
                    boxes = candidate_boxes[i]
                if boxes.shape[0] == 0:
                    continue
                if limit is not None and boxes.shape[0] > limit:
                    boxes = boxes[:limit, :]
    
                overlaps = bbox_overlaps(boxes.astype(np.float),
                                         gt_boxes.astype(np.float))
    
                _gt_overlaps = np.zeros((gt_boxes.shape[0]))
                for j in xrange(gt_boxes.shape[0]):
                    # find which proposal box maximally covers each gt box
                    argmax_overlaps = overlaps.argmax(axis=0)
                    # and get the iou amount of coverage for each gt box
                    max_overlaps = overlaps.max(axis=0)
                    # find which gt box is 'best' covered (i.e. 'best' = most iou)
                    gt_ind = max_overlaps.argmax()
                    gt_ovr = max_overlaps.max()
                    assert(gt_ovr >= 0)
                    # find the proposal box that covers the best covered gt box
                    box_ind = argmax_overlaps[gt_ind]
                    # record the iou coverage of this gt box
                    _gt_overlaps[j] = overlaps[box_ind, gt_ind]
                    assert(_gt_overlaps[j] == gt_ovr)
                    # mark the proposal box and the gt box as used
                    overlaps[box_ind, :] = -1
                    overlaps[:, gt_ind] = -1
                # append recorded iou coverage level
                gt_overlaps = np.hstack((gt_overlaps, _gt_overlaps))
    
            gt_overlaps = np.sort(gt_overlaps)
            if thresholds is None:
                step = 0.05
                thresholds = np.arange(0.5, 0.95 + 1e-5, step)
            recalls = np.zeros_like(thresholds)
            # compute recall for each iou threshold
            for i, t in enumerate(thresholds):
                recalls[i] = (gt_overlaps >= t).sum() / float(num_pos)
            # ar = 2 * np.trapz(recalls, thresholds)
            ar = recalls.mean()
            return {'ar': ar, 'recalls': recalls, 'thresholds': thresholds,
                    'gt_overlaps': gt_overlaps}
    
        def create_roidb_from_box_list(self, box_list, gt_roidb):
    #断言box_list的数目和图像数目一样，这里box_list[i]里存的是相应第i张图像里所有的bbox的坐标
            assert len(box_list) == self.num_images, \
                    'Number of boxes must match number of ground-truth images'
    #重要，roidb是一个列表，列表中的每一个元素是字典，存储了每张图象的boxes,gt_classes，'gt_overlaps'，是否flipped信息
            roidb = []
            for i in xrange(self.num_images):
                boxes = box_list[i]
                num_boxes = boxes.shape[0]
    #计算每个box和每一类目标的重叠率
                overlaps = np.zeros((num_boxes, self.num_classes), dtype=np.float32)
    
                if gt_roidb is not None and gt_roidb[i]['boxes'].size > 0:
    #取得ground-truth里bbox的坐标
                    gt_boxes = gt_roidb[i]['boxes']
    #取得每个bbox对应的类别
                    gt_classes = gt_roidb[i]['gt_classes']
    #计算roidb的bbox与ground-truth的bbox的重叠率
                    gt_overlaps = bbox_overlaps(boxes.astype(np.float),
                                                gt_boxes.astype(np.float))
    #与那一类的重叠率最大
                    argmaxes = gt_overlaps.argmax(axis=1)
                    maxes = gt_overlaps.max(axis=1)
                    I = np.where(maxes > 0)[0]
                    overlaps[I, gt_classes[argmaxes[I]]] = maxes[I]
    
                overlaps = scipy.sparse.csr_matrix(overlaps)
                roidb.append({
                    'boxes' : boxes,
                    'gt_classes' : np.zeros((num_boxes,), dtype=np.int32),
                    'gt_overlaps' : overlaps,
                    'flipped' : False,
                    'seg_areas' : np.zeros((num_boxes,), dtype=np.float32),
                })
            return roidb
    
        @staticmethod
        def merge_roidbs(a, b):
            assert len(a) == len(b)
            for i in xrange(len(a)):
                a[i]['boxes'] = np.vstack((a[i]['boxes'], b[i]['boxes']))
                a[i]['gt_classes'] = np.hstack((a[i]['gt_classes'],
                                                b[i]['gt_classes']))
                a[i]['gt_overlaps'] = scipy.sparse.vstack([a[i]['gt_overlaps'],
                                                           b[i]['gt_overlaps']])
                a[i]['seg_areas'] = np.hstack((a[i]['seg_areas'],
                                               b[i]['seg_areas']))
            return a
    
        def competition_mode(self, on):
            """Turn competition mode on or off."""
            pass

lib/datasets/pascal\_voc.py

#CODING:UTF-8
    # --------------------------------------------------------
    # Fast R-CNN
    # Copyright (c) 2015 Microsoft
    # Licensed under The MIT License [see LICENSE for details]
    # Written by Ross Girshick
    # --------------------------------------------------------
    
    import os
    from datasets.imdb import imdb
    import datasets.ds_utils as ds_utils
    import xml.etree.ElementTree as ET
    import numpy as np
    import scipy.sparse
    import scipy.io as sio
    import utils.cython_bbox
    import cPickle
    import subprocess
    import uuid
    from voc_eval import voc_eval
    from fast_rcnn.config import cfg
    
    class pascal_voc(imdb):
        def __init__(self, image_set, year, devkit_path=None):
            imdb.__init__(self, 'voc_' + year + '_' + image_set)
            self._year = year
            self._image_set = image_set
            self._devkit_path = self._get_default_path() if devkit_path is None \
                                else devkit_path
            self._data_path = os.path.join(self._devkit_path, 'VOC' + self._year)
            self._classes = ('__background__', # always index 0
                             'aeroplane', 'bicycle', 'bird', 'boat',
                             'bottle', 'bus', 'car', 'cat', 'chair',
                             'cow', 'diningtable', 'dog', 'horse',
                             'motorbike', 'person', 'pottedplant',
                             'sheep', 'sofa', 'train', 'tvmonitor')
            self._class_to_ind = dict(zip(self.classes, xrange(self.num_classes)))
            self._image_ext = '.jpg'
            self._image_index = self._load_image_set_index()
            # Default to roidb handler
            self._roidb_handler = self.selective_search_roidb
            self._salt = str(uuid.uuid4())
            self._comp_id = 'comp4'
    
            # PASCAL specific config options
            self.config = {'cleanup'     : True,
                           'use_salt'    : True,
                           'use_diff'    : False,
                           'matlab_eval' : False,
                           'rpn_file'    : None,
                           'min_size'    : 2}
    
            assert os.path.exists(self._devkit_path), \
                    'VOCdevkit path does not exist: {}'.format(self._devkit_path)
            assert os.path.exists(self._data_path), \
                    'Path does not exist: {}'.format(self._data_path)
    
        def image_path_at(self, i):
            """
            Return the absolute path to image i in the image sequence.
            """
            return self.image_path_from_index(self._image_index[i])
    
        def image_path_from_index(self, index):
            """
            Construct an image path from the image's "index" identifier.
            """
            image_path = os.path.join(self._data_path, 'JPEGImages',
                                      index + self._image_ext)
            assert os.path.exists(image_path), \
                    'Path does not exist: {}'.format(image_path)
            return image_path
    
        def _load_image_set_index(self):
            """
            Load the indexes listed in this dataset's image set file.
            """
            # Example path to image set file:
            # self._devkit_path + /VOCdevkit2007/VOC2007/ImageSets/Main/val.txt
            image_set_file = os.path.join(self._data_path, 'ImageSets', 'Main',
                                          self._image_set + '.txt')
            assert os.path.exists(image_set_file), \
                    'Path does not exist: {}'.format(image_set_file)
            with open(image_set_file) as f:
                image_index = [x.strip() for x in f.readlines()]
            return image_index
    
        def _get_default_path(self):
            """
            Return the default path where PASCAL VOC is expected to be installed.
            """
            return os.path.join(cfg.DATA_DIR, 'VOCdevkit' + self._year)
    
        def gt_roidb(self):
            """
            Return the database of ground-truth regions of interest.
    
            This function loads/saves from/to a cache file to speed up future calls.
            """
            cache_file = os.path.join(self.cache_path, self.name + '_gt_roidb.pkl')
            if os.path.exists(cache_file):
                with open(cache_file, 'rb') as fid:
                    roidb = cPickle.load(fid)
                print '{} gt roidb loaded from {}'.format(self.name, cache_file)
                return roidb
    
            gt_roidb = [self._load_pascal_annotation(index)
                        for index in self.image_index]
            with open(cache_file, 'wb') as fid:
                cPickle.dump(gt_roidb, fid, cPickle.HIGHEST_PROTOCOL)
            print 'wrote gt roidb to {}'.format(cache_file)
    
            return gt_roidb
    
        def selective_search_roidb(self):
            """
            Return the database of selective search regions of interest.
            Ground-truth ROIs are also included.
    
            This function loads/saves from/to a cache file to speed up future calls.
            """
            cache_file = os.path.join(self.cache_path,
                                      self.name + '_selective_search_roidb.pkl')
    
            if os.path.exists(cache_file):
                with open(cache_file, 'rb') as fid:
                    roidb = cPickle.load(fid)
                print '{} ss roidb loaded from {}'.format(self.name, cache_file)
                return roidb
    
            if int(self._year) == 2007 or self._image_set != 'test':
                gt_roidb = self.gt_roidb()
                ss_roidb = self._load_selective_search_roidb(gt_roidb)
                roidb = imdb.merge_roidbs(gt_roidb, ss_roidb)
            else:
                roidb = self._load_selective_search_roidb(None)
            with open(cache_file, 'wb') as fid:
                cPickle.dump(roidb, fid, cPickle.HIGHEST_PROTOCOL)
            print 'wrote ss roidb to {}'.format(cache_file)
    
            return roidb
    
        def rpn_roidb(self):
            if int(self._year) == 2007 or self._image_set != 'test':
                gt_roidb = self.gt_roidb()
                rpn_roidb = self._load_rpn_roidb(gt_roidb)
                roidb = imdb.merge_roidbs(gt_roidb, rpn_roidb)
            else:
                roidb = self._load_rpn_roidb(None)
    
            return roidb
    
        def _load_rpn_roidb(self, gt_roidb):
            filename = self.config['rpn_file']
            print 'loading {}'.format(filename)
            assert os.path.exists(filename), \
                   'rpn data not found at: {}'.format(filename)
    #得到rpn产生的box
            with open(filename, 'rb') as f:
                box_list = cPickle.load(f)
    #由box_list产生roidb
            return self.create_roidb_from_box_list(box_list, gt_roidb)
    
        def _load_selective_search_roidb(self, gt_roidb):
            filename = os.path.abspath(os.path.join(cfg.DATA_DIR,
                                                    'selective_search_data',
                                                    self.name + '.mat'))
            assert os.path.exists(filename), \
                   'Selective search data not found at: {}'.format(filename)
            raw_data = sio.loadmat(filename)['boxes'].ravel()
    
            box_list = []
            for i in xrange(raw_data.shape[0]):
                boxes = raw_data[i][:, (1, 0, 3, 2)] - 1
                keep = ds_utils.unique_boxes(boxes)
                boxes = boxes[keep, :]
                keep = ds_utils.filter_small_boxes(boxes, self.config['min_size'])
                boxes = boxes[keep, :]
                box_list.append(boxes)
    
            return self.create_roidb_from_box_list(box_list, gt_roidb)
    
        def _load_pascal_annotation(self, index):
            """
            Load image and bounding boxes info from XML file in the PASCAL VOC
            format.
            """
    #xml文件名
            filename = os.path.join(self._data_path, 'Annotations', index + '.xml')
    #解析xml
            tree = ET.parse(filename)
    #找到所有object属性项
            objs = tree.findall('object')
            if not self.config['use_diff']:
                # Exclude the samples labeled as difficult
                non_diff_objs = [
                    obj for obj in objs if int(obj.find('difficult').text) == 0]
                # if len(non_diff_objs) != len(objs):
                #     print 'Removed {} difficult objects'.format(
                #         len(objs) - len(non_diff_objs))
                objs = non_diff_objs
            num_objs = len(objs)
    #boxes存储这张图片上所有bbox的坐标
            boxes = np.zeros((num_objs, 4), dtype=np.uint16)
    #gt_classes存储每个bbox的类别
            gt_classes = np.zeros((num_objs), dtype=np.int32)
            overlaps = np.zeros((num_objs, self.num_classes), dtype=np.float32)
            # "Seg" area for pascal is just the box area
            seg_areas = np.zeros((num_objs), dtype=np.float32)
    
            # Load object bounding boxes into a data frame.
            for ix, obj in enumerate(objs):
                bbox = obj.find('bndbox')
                # Make pixel indexes 0-based
                x1 = float(bbox.find('xmin').text) - 1
                y1 = float(bbox.find('ymin').text) - 1
                x2 = float(bbox.find('xmax').text) - 1
                y2 = float(bbox.find('ymax').text) - 1
    #从类别名字映射到ID
                cls = self._class_to_ind[obj.find('name').text.lower().strip()]
                boxes[ix, :] = [x1, y1, x2, y2]
                gt_classes[ix] = cls
    #因为是groud-truth，所以重叠率设置为1
                overlaps[ix, cls] = 1.0
                seg_areas[ix] = (x2 - x1 + 1) * (y2 - y1 + 1)
    
            overlaps = scipy.sparse.csr_matrix(overlaps)
    #返回一个字典
            return {'boxes' : boxes,
                    'gt_classes': gt_classes,
                    'gt_overlaps' : overlaps,
                    'flipped' : False,
                    'seg_areas' : seg_areas}
    
        def _get_comp_id(self):
            comp_id = (self._comp_id + '_' + self._salt if self.config['use_salt']
                else self._comp_id)
            return comp_id
    
        def _get_voc_results_file_template(self):
            # VOCdevkit/results/VOC2007/Main/<comp_id>_det_test_aeroplane.txt
            filename = self._get_comp_id() + '_det_' + self._image_set + '_{:s}.txt'
            path = os.path.join(
                self._devkit_path,
                'results',
                'VOC' + self._year,
                'Main',
                filename)
            return path
    
        def _write_voc_results_file(self, all_boxes):
            for cls_ind, cls in enumerate(self.classes):
                if cls == '__background__':
                    continue
                print 'Writing {} VOC results file'.format(cls)
                filename = self._get_voc_results_file_template().format(cls)
                with open(filename, 'wt') as f:
                    for im_ind, index in enumerate(self.image_index):
                        dets = all_boxes[cls_ind][im_ind]
                        if dets == []:
                            continue
                        # the VOCdevkit expects 1-based indices
                        for k in xrange(dets.shape[0]):
                            f.write('{:s} {:.3f} {:.1f} {:.1f} {:.1f} {:.1f}\n'.
                                    format(index, dets[k, -1],
                                           dets[k, 0] + 1, dets[k, 1] + 1,
                                           dets[k, 2] + 1, dets[k, 3] + 1))
    
        def _do_python_eval(self, output_dir = 'output'):
            annopath = os.path.join(
                self._devkit_path,
                'VOC' + self._year,
                'Annotations',
                '{:s}.xml')
            imagesetfile = os.path.join(
                self._devkit_path,
                'VOC' + self._year,
                'ImageSets',
                'Main',
                self._image_set + '.txt')
            cachedir = os.path.join(self._devkit_path, 'annotations_cache')
            aps = []
            # The PASCAL VOC metric changed in 2010
            use_07_metric = True if int(self._year) < 2010 else False
            print 'VOC07 metric? ' + ('Yes' if use_07_metric else 'No')
            if not os.path.isdir(output_dir):
                os.mkdir(output_dir)
            for i, cls in enumerate(self._classes):
                if cls == '__background__':
                    continue
                filename = self._get_voc_results_file_template().format(cls)
                rec, prec, ap = voc_eval(
                    filename, annopath, imagesetfile, cls, cachedir, ovthresh=0.5,
                    use_07_metric=use_07_metric)
                aps += [ap]
                print('AP for {} = {:.4f}'.format(cls, ap))
                with open(os.path.join(output_dir, cls + '_pr.pkl'), 'w') as f:
                    cPickle.dump({'rec': rec, 'prec': prec, 'ap': ap}, f)
            print('Mean AP = {:.4f}'.format(np.mean(aps)))
            print('~~~~~~~~')
            print('Results:')
            for ap in aps:
                print('{:.3f}'.format(ap))
            print('{:.3f}'.format(np.mean(aps)))
            print('~~~~~~~~')
            print('')
            print('--------------------------------------------------------------')
            print('Results computed with the **unofficial** Python eval code.')
            print('Results should be very close to the official MATLAB eval code.')
            print('Recompute with `./tools/reval.py --matlab ...` for your paper.')
            print('-- Thanks, The Management')
            print('--------------------------------------------------------------')
    
        def _do_matlab_eval(self, output_dir='output'):
            print '-----------------------------------------------------'
            print 'Computing results with the official MATLAB eval code.'
            print '-----------------------------------------------------'
            path = os.path.join(cfg.ROOT_DIR, 'lib', 'datasets',
                                'VOCdevkit-matlab-wrapper')
            cmd = 'cd {} && '.format(path)
            cmd += '{:s} -nodisplay -nodesktop '.format(cfg.MATLAB)
            cmd += '-r "dbstop if error; '
            cmd += 'voc_eval(\'{:s}\',\'{:s}\',\'{:s}\',\'{:s}\'); quit;"' \
                   .format(self._devkit_path, self._get_comp_id(),
                           self._image_set, output_dir)
            print('Running:\n{}'.format(cmd))
            status = subprocess.call(cmd, shell=True)
    
        def evaluate_detections(self, all_boxes, output_dir):
            self._write_voc_results_file(all_boxes)
            self._do_python_eval(output_dir)
            if self.config['matlab_eval']:
                self._do_matlab_eval(output_dir)
            if self.config['cleanup']:
                for cls in self._classes:
                    if cls == '__background__':
                        continue
                    filename = self._get_voc_results_file_template().format(cls)
                    os.remove(filename)
    
        def competition_mode(self, on):
            if on:
                self.config['use_salt'] = False
                self.config['cleanup'] = False
            else:
                self.config['use_salt'] = True
                self.config['cleanup'] = True
    
    if __name__ == '__main__':
        from datasets.pascal_voc import pascal_voc
        d = pascal_voc('trainval', '2007')
        res = d.roidb
        from IPython import embed; embed()

http://blog.csdn.net/iamzhangzhuping/article/category/6230157

http://blog.csdn.net/u010668907/article/category/6237110

[Faster RCNN代码理解（Python）][Faster RCNN_Python]

[Faster RCNN_Python]: http://blog.csdn.net/u011956147/article/details/53053381