torchvision

torchvision是独立于pytorch的关于图像操作的一些方便工具库。

pytorch发布了torchvision0.3, 里面实现了Mask_RCNN, Keypoint_RCNN和DeepLabV3，可以直接用于语义分割，目标检测，实例分割和人体关键点检测4个任务。

在github上torch/vision/reference里面有classification, detection和segmentation三个文件夹，分别对应不同任务。直接运行detection的代码是MaskRCNN的实现，用于目标检测和实例分割任务。官网也有对应的教程。可以轻松通过COCO2017数据集进行训练和测试。

安装

参考：PyTorch安装

Dataset

下载COCO2017数据集，下载train2017，val2017和annotations三个文件后解压，最终文件目录结构如下

data/COCO2017/

​ train2017/

​ val2017/

​ annotations/

演示

下载：pytorch_torchvision_demo.zip

人体关键点演示：

修改参数detect_threshold和keypoint_score_threshold, 分别过滤得分低的个体和得分低的关键点。

运行python predict_keypoint_visualize.py

Mask_RCNN演示：

运行python predict_maskrcnn_visualize.py

在根目录下建立文件夹result, 可视化后的图片存在在此文件夹下。

Train

对train.py进行如下操作

  • 修改函数get_dataset中的paths

  • 修改文件中列出了的各种参数

  • 假设路径都修改完毕，使用预训练模型进行训练: python train.py --pretrained

Evaluate

执行代码python train.py --test-only

程序在COCO2017 val集上的结果如下，同官网介绍一致

  1. Averaged stats: model_time: 0.1371 (0.1627)  evaluator_time: 0.0043 (0.0105)
  2. Accumulating evaluation results...
  3. DONE (t=1.31s).
  4. Accumulating evaluation results...
  5. DONE (t=0.41s).
  6. IoU metric: bbox
  7.  Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.502
  8.  Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.796
  9.  Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.545
  10.  Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.341
  11.  Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.591
  12.  Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.648
  13.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.176
  14.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.519
  15.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.603
  16.  Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.460
  17.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.669
  18.  Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.738
  19. IoU metric: keypoints
  20.  Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets= 20 ] = 0.599
  21.  Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets= 20 ] = 0.834
  22.  Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets= 20 ] = 0.650
  23.  Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.553
  24.  Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.675
  25.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 20 ] = 0.672
  26.  Average Recall     (AR) @[ IoU=0.50      | area=   all | maxDets= 20 ] = 0.889
  27.  Average Recall     (AR) @[ IoU=0.75      | area=   all | maxDets= 20 ] = 0.721
  28.  Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.623
  29.  Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.741