Unsupervised Domain Adaptive Re-Identification

Implementation of the paper Unsupervised Domain Adaptive Re-Identification: Theory and Practice.

The selftraining scheme proposed in the paper is simple yet effective.

Setup

1. Datasets (source dataset and target dataset).
2. Pre-trained (on source dataset) model.

Requirements

• PyTorch

Running the experiments

To replicate the results in the paper, you can download pre-trained models on Market1501 from GoogleDrive and on DukeMTMC from GoogleDrive. Our models are trained with PyTorch 0.3.

python selftraining.py \
    --src_dataset <name_of_source_dataset>\
    --tgt_dataset <name_of_target_dataset>\
    --resume <dir_of_source_trained_model>\
    --data_dir <dir_of_source_target_data>\
    --logs_dir <dir_to_save_model_after_adaptation>

dw_example.ipynb is the file for replicating Figure 6 in the paper.

Results

DukeMTMC ---> Market1501

	Rank-1	Rank-5	Rank-10	mAP
On source (DukeMTMC)	80.8	91.2	94.2	65.4
On target (Market1501)	46.8	64.6	71.5	19.1
After adaptation	75.8	89.5	93.2	53.7

Market1501 ---> DukeMTMC

	Rank-1	Rank-5	Rank-10	mAP
On source (Market1501)	91.6	97.1	98.5	78.2
On target (DukeMTMC)	27.3	41.2	47.1	11.9
After adaptation	68.4	80.1	83.5	49.0

Acknowledgement

Our code is based on open-reid.