mmpose

mmpose It's another easy-to-use tool box under Shangtang , It is mainly developed for attitude recognition task , Including human posture recognition and animal posture recognition . It inherits many ready-made networks for everyone to use , And provides trained weights . Users can train their data with their own needs , Or directly apply their results , Anyway, it's very comprehensive ！！！！
github Connect ：https://github.com/open-mmlab/mmpose
Use connections ：https://mmpose.readthedocs.io/en/latest/demo.html

Installation tutorial

1. install mmcv-full

Suppose your computer has installed anaconda Environmental Science , And you've installed pytorch, first-class pytorch The version is 1.5 above , If not , It is best to create a new virtual environment using the following code , I personally like the latest version , Because the compatibility of current installation packages is relatively good , If you don't understand, don't worry about new compatibility problems , The new version can basically run the old version of the code , You can rest assured , I don't say much nonsense , Let's start .
pytorch Version less than 1.5 Use the following code ：
Pay attention to your cuda Version ha , I came according to my own , see cuda edition , Available on the command line nvcc -V

conda create -n open-mmlab python=3.8 pytorch=1.9 cudatoolkit=11.3 torchvision -c pytorch -y
conda activate open-mmlab

Now it has been satisfied pytorch The version is greater than 1.5 了 , Let's start installing openmim
Pay attention to your version number , Change to the corresponding , If you don't want to change , Use directly pip install mmcv-full Save trouble

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.9.0/index.html

2. install MMPose

Here's the advice win Students of the system , because win No, linux Command line, you're so comfortable , I suggest you install one git cmd, This is very cool to use , Can let you again linux The feeling of . Okay , Start installation mmpose 了
clone mmpose library

git clone https://github.com/open-mmlab/mmpose.git
cd mmpose

Install required packages

pip install -r requirements.txt
pip install -v -e .  # or "python setup.py develop"

Note here that during installation , Some packages may not be installed , At this time, you need to install it manually , I had a problem during the installation poseval This library , It can't be installed , I searched Baidu for the following , Here is the installation method for you .
It can't be installed poseval, Use the following manual method to install
1. The first way ：

pip install https://github.com/svenkreiss/poseval.git

2. The second way ：

git clone https://github.com/svenkreiss/poseval.git
cd poseval
pip install -e .  # install the local package ('.') in editable mode ('-e')

Note that after a problem occurs halfway , Some packages in the back are not installed properly , You also need to use pip install -r requirements.txt continue installation

2. install mmdet

This can be installed or not , I suggest you install the following , Because the steps are simple , If you don't install it , Later, the library of attitude estimation is used top-down The idea needs to mark the position of the box , Do it yourself json file , trouble , It's easy to install here .

pip install mmdet

Use the tutorial

1. Forecast picture

After all the above are installed , Now let's start using it
If we don't install mmdet word , Only test the pictures they provide
Now let's coco Data sets, for example , Test and play this library , They provide marked json

python demo/top_down_img_demo.py configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_256x192.py  https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_256x192-b9e0b3ab_20200708.pth  --img-root tests/data/coco/ --json-file tests/data/coco/test_coco.json --out-img-root vis_results

Here are the predicted results

If you install it mmdet word , Any picture can predict
Use the following code directly ,mmdet You can detect people by yourself , There is no need to mark manually

python demo/top_down_img_demo_with_mmdet.py demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth  configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_256x192.py  https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_256x192-b9e0b3ab_20200708.pth --img-root tests/data/coco/ --img 1.jpeg --out-img-root vis_results

Let's take a look at the effect

2. Prediction video

The code for predicting video is as follows ：

python demo/top_down_video_demo_with_mmdet.py demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py  https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_256x192.py https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_256x192-b9e0b3ab_20200708.pth --video-path demo/resources/1.mp4 --out-video-root vis_results

Let's take a look at the effect

Last , If you need to recognize animals, you can see the following tutorial
https://mmpose.readthedocs.io/en/latest/demo.html