Pytorch implementation of "Fast Training of Triplet-based Deep Binary Embedding Networks". http://arxiv.org/abs/1603.02844

Feel free to contribute code.

Refactor this project.

Use code in https://github.com/kentsommer/keras-inceptionV4 to extract feature.

Deep hash for "A", "B".

• Add multiclass support.
• Make code clean.
• Add more base networks.
• Add query code for new project.

1. Put training pictures in train/[category-id], test pictures in data/test.
2. Run src/extract_feature/batch_extarct_test.py and src/extract_feature/batch_extract_train.py to extract feature for future use.
3. Run src/hash_net/generate_random_dataset.py to generate random training data.
4. Run src/hash_net/hashNet.py to train your triplet deep hash network.

## Test

1. Create folder test, and create pos, neg in test with pictures that you want to retrive.

2. Run testQue.py to query your picture set.