Skip to content

Latest commit

 

History

History
63 lines (42 loc) · 2.2 KB

README.md

File metadata and controls

63 lines (42 loc) · 2.2 KB

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.

Illustration of the selftraining scheme.

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.