This is the code for the paper:
Hybrid Style Siamese Network: Incorporating style loss in complementary apparels retrieval
Mayukh Bhattacharyya, Sayan Nag
Computer Vision for Fashion, Art and Design, CVPR 2020
Hybrid Style Siamese Network incorporates style loss into triplet loss, in order to aid in complementary images retrieval. In the paper, it had been used for the application of complementary apparels retrieval.
The performance in regards to the retrieval of complementary items in terms of MAP scores.
| Model | Seed 1 | Seed 2 | Seed 3 | Mean |
|---|---|---|---|---|
| Siamese Network | 0.1226 | 0.1323 | 0.1263 | 0.1271 |
| Hybrid Style Siamese Network | 0.1251 | 0.1343 | 0.1329 | 0.1308 |
- pytorch (both torch and torchvision)
- numpy
- opencv
- scikit-learn
In order to reproduce the experiments in the paper:
python hssn.py -s seed_value -l learning rate -e epochs -p patience -b batch_size -f cv_fold_id --hybrid
--hybrid is the flag for running the hybrid style siamese network. Omitting it will run the same experiment with the normal siamese network. Explanation of other parameters are there in hssn.py