Facebook AI Image Similarity Challenge: Descriptor Track

This repository contains the code for our solution to the Facebook AI Image Similarity Challenge: Descriptor Track, hosted on DrivenData:
https://www.drivendata.org/competitions/85/competition-image-similarity-2-final/leaderboard/

The solution is described in the paper hosted on arXiv.org:
https://arxiv.org/abs/2112.03415v1

Environment Setup

To run the models, you must first configure the environment:

• Download the datasets following the README.md instructions at: all_datasets/README.md

• Copy some query images. Go to the directory FB_page_qry/ and run: python3 copy_fb_query_images.py

Checkpoints

Download all the checkpoints from my gdrive and locate them inside checkpoints/ directory.

Phase 1 inference

Follow the README.md instructions at: phase1_scripts/README.md

Phase 2 inference

Follow the README.md instructions at: phase2_scripts/README.md

Training of Final Ensemble Models

Follow the README.md instructions at: ensemble_training_scripts/README.md

Training your model

Follow the README.md instructions at: training_scripts/README.md