Supplementary List

1. src, model training, post processing and evaluation code
2. data, gallery and query features using SAKE model with distribution alignment loss

Code Usage

1. (In the repo root dir) Calculate and save scores.

   python src/pre_process.py
   

2. Run cluster and retrieval experiments。

• KMeans w/ fuse

  python src/clusterretr.py --feature_file ./data/features_processed.pickle > results/kld_sketchy_fuse0.2.log --fuse 0.2
  

• KMeans w/o fuse

  python src/clusterretr.py --feature_file ./data/features_processed.pickle > results/kld_sketchy.log