• 项目中文名
• 作者
• 项目简介
• • 1. 功能
  • 2. 性能
  • 3. 评估指标
  • 4. 使用数据集
• 运行环境与依赖
• 输入与输出
• 运行方式

项目中文名

以摄像机为基本单位的批归一化技术用于行人图像再识别特征提取

Camera-based Batch Normalization for Person Re-identification

作者

Yongfei Zhang, Tianyu Zhang

项目简介

1. 功能

This repo has the training and testing codes of supervised learning with Camera-based batch normalization Rethinking the Distribution Gap of Person Re-identification with Camera-based Batch Normalization. It aligns camera distributions of both training cameras and testing cameras.

We design a demo for testing.

2. 性能

The training phase uses about 10G of GPU memory.

The ReID accuarcy can be seen in CBN.

3. 评估指标

Rank-1 accuracy and mAP.

4. 使用数据集

Market-1501 , DukeMTMC-reID and MSMT17 are used.

运行环境与依赖

torch==1.3.1
torchvision==0.4.2
tensorboard
future
fire
tqdm

输入与输出

Name	Notes
Input	The file path of one person image.
Output	An array containing the feature vector.

运行方式

1. Train

CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=0,1 \
python train_model.py train --trainset_name market --save_dir='market_demo' --max_epoch 60 --decay_epoch 40

2. Evaluate

CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=0 \
python test_model.py test --testset_name market --save_dir='market_demo'

You may refer to config.py to adjust more hyper-parameters.

The demo code is implemented in cbn_demo.py. We write a new class for users to call.