1. Introduction to fastreid project
Code address: https://github.com/JDAI-CV/fast-reid
The project is open sourced by Jingdong AI Technology Research Institute, and the main research direction is ReID, including pedestrian ReID and other ReID.
The blogger has been using this project to study pedestrian ReID and vehicle ReID, and has achieved good results on private data sets. But now it is introduced to use this project for pedestrian attribute recognition, and the data set used is the pedestrian data set PA100K.
2. Training pedestrian attribute recognition
(1) Download the fastreid project: git clone https://github.com/JDAI-CV/fast-reid.git
(2) Download pedestrian data set PA100K, download address: download and analyze pedestrian attribute data set PA100K
(3) Place the image data and annotation files in the datasets/ directory of the project, as shown in the following figure:
Note: Pictures are stored in data.
(4) Run the following command to start training the model (before training the model, modify the path of the pre-training model yourself)
python projects/attribute_recognition/train_net.py --config-file projects/attribute_recognition/configs/pa100.yml
The effect of starting training is as follows:
(5) The test results after training are as follows
The above is just a baseline, but the test index has exceeded the index of many papers. There is room for improvement in various test indicators after tuning.
For technical exchanges, please leave a message or WeChat: