deeplab v3+ 代码复现简单官方 voc2012

梦里梦外; 2024-04-17 06:28 74阅读 0赞

# 为了使local\_test.sh顺利运行。 #

代码来源于官方，就是论文作者：

[https://github.com/tensorflow/models/tree/master/research/deeplab][https_github.com_tensorflow_models_tree_master_research_deeplab]

那肯定是需要先下载代码的。

# 1.安装slim工具包。很简单 #

只需要两步，下载，安装。必须安装不然import net 会报错

[https://blog.csdn.net/u013249853/article/details/100089276][https_blog.csdn.net_u013249853_article_details_100089276]

# 2.手工准备好数据库文件，或者自动下载。 #

总共包含两个文件，一个是数据，voc，另一个是checkpoint。下面两种准备方式，请在情况的允许下使用自动方式。

## 手工准备 ##

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70][]

需要在deeplab文件夹下准备这两个文件第一个是VOCdevkit，里面是官方的trainval包解压

"VOCtrainval_11-May-2012.tar"解压

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70 1][]

第二个是从[https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model\_zoo.md][https_github.com_tensorflow_models_blob_master_research_deeplab_g3doc_model_zoo.md]下载的

[http://download.tensorflow.org/models/deeplabv3\_pascal\_train\_aug\_2018\_01\_04.tar.gz][http_download.tensorflow.org_models_deeplabv3_pascal_train_aug_2018_01_04.tar.gz]点击这个应该就行

或者

<table> 
 <tbody> 
  <tr> 
   <td><a href="http://download.tensorflow.org/models/deeplabv3_pascal_train_aug_2018_01_04.tar.gz" rel="nofollow">xception65_coco_voc_trainaug</a></td> 
   <td>16&nbsp;<br> 8</td> 
   <td>[1.0]&nbsp;<br> [0.5:0.25:1.75]</td> 
   <td>No&nbsp;<br> Yes</td> 
   <td>54.17B&nbsp;<br> 3055.35B</td> 
   <td>0.7&nbsp;<br> 223.2</td> 
   <td>82.20% (val)&nbsp;<br> 83.58% (val)</td> 
   <td>439MB</td> 
  </tr> 
 </tbody> 
</table>

放到下图所示位置

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70 2][]

local\_test.py中：

sh download_and_convert_voc2012.sh

负责下载voc数据

wget -nd -c "${TF_INIT_ROOT}/${TF_INIT_CKPT}"

负责下载模型压缩包

所以你需要将上面两行注释掉要不他还是会下载。

**手工主要是考虑到网速问题。**

**这里必须提一下，如果提示你checkpoint有问题，建议删除checkpoint文件也就是这里的init\_model包，并且执行自动下载。**

## 自动下载 ##

直接运行local\_test.py即可。

之后就是输出train.py结果：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70 3][]

另外如果不训练，那么将iteration 也就是local\_test.sh里面的

NUM_ITERATIONS=10

改成1000次什么的

保存的variable值将会在：.../deeplab/datasets/pascal\_voc\_seg/exp/train\_on\_trainval\_set/train

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70 4][]

[https_github.com_tensorflow_models_tree_master_research_deeplab]: https://github.com/tensorflow/models/tree/master/research/deeplab
[https_blog.csdn.net_u013249853_article_details_100089276]: https://blog.csdn.net/u013249853/article/details/100089276
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/17/8b6f37c5d7ec43f4b71c2cb2932b3d63.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70 1]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/17/93037e44b7734690b43748c3d796e71e.png
[https_github.com_tensorflow_models_blob_master_research_deeplab_g3doc_model_zoo.md]: https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md
[http_download.tensorflow.org_models_deeplabv3_pascal_train_aug_2018_01_04.tar.gz]: http://download.tensorflow.org/models/deeplabv3_pascal_train_aug_2018_01_04.tar.gz
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70 2]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/17/36603eea22d84003a3385a504479142a.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70 3]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/17/c19a16784b0949f78654a61e7ef8854b.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTMyNDk4NTM_size_16_color_FFFFFF_t_70 4]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/17/ff912fb9f4a9431d82c0cfe51a792352.png