Computer Vision ToolBox of Pytorch

• ObjectDection Human Bbox Detection

  • YoloV3
  • RFCN
  • CornerNet

• 2D pose estimation (kps2d__detection)

  • Simple BaseLine
  • High Resolution Network (hr_net)
  • Open Pose

• 3D pose estimation (kps3d_detection)

  • VideoPose3D

• Object Tracking

  • lighttrack
  • pose flow

• ActionRecgnition

  • ST-GCN
  • 2S-AGCN

Install

Install pytorch >= v1.0.0 pip install -r requirment.txt

Make

cd kps2d_detection/hr_net/lib/ make

cd pose_track/lighttrack/graph/torchlight python setup.py install

add Environment Variable

vim ~/.bashrc

export PATH=/path/to/cvToolBox:$PATH

export PYTHONPATH=/path/to/cvToolBox:$PYTHONPATH

Example of Human Bbox Detection

```
from object_detection.yolo_v3.human_detector import load_model
from object_detection.yolo_v3.human_detector import inference
import cv2
im = cv2.imread('data/test.png')
model = load_model()
# return bbox and its probability
inference(im, model)
```

Example of pose track

cd pose_track/lighttrack/

• python demo.py -p 0 (pose estimator is simple-baseline, pose speed 90 person/s)

• python demo.py -p 1 (pose estimator is hrnet, pose speed 25 person/s)