AICity Challenge 2020 Track2 vehicle re-identification(re-id).

This branch contains paddlepaddle training codes for aicity 2020 Track2 vehicle re-identification. More detail can be found in aicity

Two architectures of feature net are supported:

single branch(traditional reid model)
multi branch

Besides, multi-task learning (reid, color, type) is also supported

Imagenet pretrained models can be found in PaddleCV

1. Environment installation

CUDA 9
cuDNN >=7.3
paddlepaddle-gpu == 1.6.3

2. Process aicity 2020 re-id data

tools are put in process_aicity_data directory, please see README in this directory

3. Training

modify configurations in config.py or arguments in train.sh
sh train.sh

4 Testing

modify configurations in config.py or arguments in test.sh
sh test.sh