Close-set Camera Style Distibution Align for Single Camera Person Re-identification

Environment

The code has been tested on Pytorch 1.1.0 and Python 3.6.

Install other required packages

pip install -r requirements.txt

Note: Our code is only tested with Python3.

We use ResNet-50 as the backbone. A pretrained model file is needed. Please put this file in the reid/weights/pre_train directory.

Dataset Preparation

1. Download Market-SCT BaiduYun (password: 1234) and Duke-SCT BaiduYun (password: 1234)

2. Make new directories in data and organize them as follows:

+-- data
|   +-- market_sct
|       +-- bounding_box_train_sct
|       +-- query
|       +-- boudning_box_test
|   +-- duke_sct
|       +-- bounding_box_train_sct
|       +-- query
|       +-- boudning_box_test

3. Train with our Proposed CCSDA.

Generate style transfer images

Train the CycleGAN-for-Camstyle to generate style transfer images and then add them to the training set of the SCT datasets

Train and test

To train with our proposed CCSDA, simply run train_bm.sh.

To evaluate trained models, simply run test_bm.sh with single GPU.

Note: We conducted all our experiments on single Tesla V100 GPU. Using multi GPU training models may cause performance degradation.