ABD-Net: Attentive but Diverse Person Re-Identification

Code for this paper ABD-Net: Attentive but Diverse Person Re-Identification

Tianlong Chen, Shaojin Ding*, Jingyi Xie*, Ye Yuan, Wuyang Chen, Yang Yang, Zhou Ren, Zhangyang Wang

In ICCV 2019

Refer to Training Guides README here, original README here, datasets README here, Model ZOO README here.

We provide complete usage pretrained models for our paper.

• Market1501 best ABD-Net model
• Duke best ABD-Net model
• MSMT17 best ABD-Net model

More models will come soon. If you want a pretrained model for some specific datasets, please be free to post an issue in our repo.

Overview

Attention mechanism has been shown to be effective for person re-identification (Re-ID). However, the learned attentive feature embeddings which are often not naturally diverse nor uncorrelated, will compromise the retrieval performance based on the Euclidean distance. We advocate that enforcing diversity could greatly complement the power of attention. To this end, we propose an Attentive but Diverse Network (ABD-Net), which seamlessly integrates attention modules and diversity regularization throughout the entire network, to learn features that are representative, robust, and more discriminative.

Here are the visualization of attention maps. (i) Original images; (ii) Attentive feature maps; (iii) Attentive but diverse feature maps. Diversity can be observed to make attention "broader" in general, and to correct some mistaken over-emphasis (such as clothes textures) by attention. (L: large values; S: small values.)

Methods

We add a CAM (Channel Attention Module) and O.F. on the outputs of res_conv_2 block. The regularized feature map is used as the input of res_conv_3. Next, after the res_conv_4 block, the network splits into a global branch and an attentive branch in parallel. We apply O.W. on all conv layers in our ResNet-50 backbone, i.e.​, from res_conv_1 to res_conv_4 and the two res_conv_5 in both branches. The outputs of two branches are concatenated as the final feature embedding.

Here are the detailed structures of CAM (Channel Attention Module) and PAM (Position Attention Module).

Results

Our proposed ABD-Net achieves the state-of-the-art (SOTA) performance in Market-1501, DukeMTMC-Re-ID and MSMT17 datasets. The detailed comparison with previous SOTA can be found in our paper.

Here are three Re-ID examples of ABD-Net (XE), Baseline + PAM + CAM and Baseline on Market-1s501. Left: query image. Right: i): top-5 results of ABD-Net (XE). ii): top-5 results of Baseline + PAM + CAM. iii): top-5 results of Baseline. Images in red boxes are negative results.

Citation

If you use this code for your research, please cite our paper.

​```
@InProceedings{Chen_2019_ICCV,
author = {Tianlong Chen and Shaojin Ding and Jingyi Xie and Ye Yuan and Wuyang Chen and Yang Yang and Zhou Ren and Zhangyang Wang},
title = {ABD-Net: Attentive but Diverse Person Re-Identification},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {Oct},
year = {2019}
}
​```