This repository is dedicated for my term paper after the 3rd year on the topic "Triplet loss modifications for deep metric learning tasks".
All code of experiments can be found here. This code uses Python language, PyTorch and Open Metric Learning libraries with all their dependecies. All requirements can be seen in requirements.txt.
All experiments are made on CUB-200-2011 dataset which contains 11788 images of 200 subcategories belonging to birds (5994 for training and 5794 for testing).
Example of code for model fine-tuning which I used for making experiments you can find in tp3-cub.ipynb notebook.
All experiments were logged in wandb.