Model training

Model is defined in representation.ipynb notebook. Hyperparameters are defined at the top of the notebook.

Implemented methods

• triplet_loss_batch_all - uses all valid triplets in given batch
• triplet_loss_batch_hard_negative - for every anchor-positive pair only the hardest negative is used (the closest one)
• triplet_loss_batch_hard - for every anchor uses only hardest postive and hardest negative

Dependencies

tensorflow==1.13

Changelog

01-04-2019

• added comments to notebook
• removed redundant code