Cross-Modal-Re-ID-baseline

Pytorch Code for Cross-Modality Person Re-Identification (Visible Thermal Re-ID) on RegDB dataset [1] and SYSU-MM01 dataset [2].

We adopt the two-stream network structure introduced in [3]. ResNet50 is adopted as the backbone. The softmax loss is adopted as the baseline.

Datasets	Pretrained	Rank@1	mAP	Model
#RegDB	ImageNet	~ 22.4%	~ 22.8%	-----
#SYSU-MM01	ImageNet	~ 24.5%	~ 27.2%	GoogleDrive

*Both of these two datasets may have some fluctuation due to random spliting. The results might be better by finetuning the hyper-parameters.

1. Prepare the datasets.

(1) RegDB Dataset [1]: The RegDB dataset can be downloaded from this website by submitting a copyright form.
- (Named: "Dongguk Body-based Person Recognition Database (DBPerson-Recog-DB1)" on their website).
- A private download link can be requested via sending me an email (mangye16@gmail.com).
(2) SYSU-MM01 Dataset [2]: The SYSU-MM01 dataset can be downloaded from this website.
- run python pre_process_sysu.py to pepare the dataset, the training data will be stored in ".npy" format.

2. Training.

Train a model by

python train.py --dataset sysu --lr 0.01 --drop 0.0 --trial 1 --gpu 1

--dataset: which dataset "sysu" or "regdb".
--lr: initial learning rate.
--drop: dropout ratio.
--trial: training trial (only for RegDB dataset).
--gpu: which gpu to run.

You may need mannully define the data path first.

Parameters: More parameters can be found in the script.

Sampling Strategy: N (= bacth size) person identities are randomly sampled at each step, then randomly select one visible and one thermal image. Details can be found in Line 302-307 in train.py.

Training Log: The training log will be saved in log/" dataset_name"+ log. Model will be saved in save_model/.

3. Testing.

Test a model on SYSU-MM01 or RegDB dataset by

python test.py --mode all --resume 'model_path' --gpu 1 --dataset sysu

--dataset: which dataset "sysu" or "regdb".
--mode: "all" or "indoor" all search or indoor search (only for sysu dataset).
--trial: testing trial (only for RegDB dataset).
--resume: the saved model path.
--gpu: which gpu to run.

4. Tips.

Softmax loss is not good on RegDB dataset, triplet loss performs much better on this dataset.

5. References.

[1] D. T. Nguyen, H. G. Hong, K. W. Kim, and K. R. Park. Person recognition system based on a combination of body images from visible light and thermal cameras. Sensors, 17(3):605, 2017.

[2] A. Wu, W.-s. Zheng, H.-X. Yu, S. Gong, and J. Lai. Rgb-infrared crossmodality person re-identification. In IEEE International Conference on Computer Vision (ICCV), pages 5380–5389, 2017.

[3] M. Ye, Z. Wang, X. Lan, and P. C. Yuen. Visible thermal person reidentification via dual-constrained top-ranking. In International Joint Conference on Artificial Intelligence (IJCAI), pages 1092–1099, 2018.