Skip to content
[WACV 2020] "Calibrated Domain-Invariant Learning for Highly Generalizable Large Scale Re-Identification" by Ye Yuan, Wuyang Chen, Tianlong Chen, Yang Yang, Zhou Ren, Zhangyang Wang, Gang Hua
Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
adin
baseline
figures update Dec 18, 2019
.gitignore
README.md
argument.py

README.md

Calibrated Domain-Invariant Learning for Highly Generalizable Large Scale Re-Identification

Calibrated Domain-Invariant Learning for Highly Generalizable Large Scale Re-Identification

Ye Yuan, Wuyang Chen, Tianlong Chen, Yang Yang, Zhou Ren, Zhangyang Wang, Gang Hua

In WACV 2020.

Overview

Many real-world applications, such as city-scale traffic monitoring and control, requires large-scale re-identification. However, previous ReID methods often failed to address two limitations in existing ReID benchmarks, i.e., low spatiotemporal coverage and sample imbalance. Notwithstanding their demonstrated success in every single benchmark, they have difficulties in generalizing to unseen environments. As a result, these methods are less applicable in a large-scale setting due to poor generalization.

In seek for a highly generalizable large-scale ReID method, we present an adversarial domain invariant feature learning framework (ADIN) that explicitly learns to separate identity-related features from challenging variations, where for the first time "free" annotations in ReID data such as video timestamp and camera index are utilized.

Furthermore, we find that the imbalance of nuisance classes jeopardizes the adversarial training, and for mitigation we propose a calibrated adversarial loss that is attentive to nuisance distribution. Experiments on existing large-scale person vehicle ReID datasets demonstrate that ADIN learns more robust and generalizable representations, as evidenced by its outstanding direct transfer performance across datasets, which is a criterion that can better measure the generalizability of large-scale ReID methods.

Methods

ADIN
Adversarial Model

ADIN
Dual Branch Structure

Training

Please sequentially finish the following steps:

  1. python baseline/script/train.py --dataset MSMT17 --loss crossEntropy (save checkpoint)
  2. python baseline/script/train.py --dataset MSMT17 --loss classCamIdAndTimeStamp --resume-checkpoint checkpoint (use the saved checkpoint)
  3. python adin/fusePretrain.py --resume-pid checkpoint_1 --resume-env checkpoint_2
  4. python adin/adinTrain.py --dataset MSMT17 --loss crossEntropy-classCamIdAndTimeStamp --resume-checkpoint checkpoint
  5. python adin/splitModel.py --checkpoint checkpoint

Evaluation

Run script

  1. python script/featureExtract.py
  2. python script/evaluate.py

Citation

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

@inproceedings{yuan2020ADIN,
  title={Collaborative Global-Local Networks for Memory-Efficient Segmentation of Ultra-High Resolution Images},
  author={Ye Yuan, Wuyang Chen, Tianlong Chen, Yang Yang, Zhou Ren, Zhangyang Wang, Gang Hua},
  booktitle={WACV},
  year={2020}
}
You can’t perform that action at this time.