2nd Place Solution to Facebook AI Image Similarity Challenge Matching Track

TL;DR

This paper describes a high-performing image similarity solution using self-supervised learning and Vision Transformers, achieving second place in a Facebook AI challenge by concatenating image pairs for direct prediction.

Contribution

The novel approach of concatenating query and reference images and training ViT to predict their relationship is a key innovation in image similarity tasks.

Findings

Achieved 0.8291 Micro-average Precision on private leaderboard

Utilized self-supervised learning with Vision Transformer for image matching

Outperformed many existing methods in the challenge

Abstract

This paper presents the 2nd place solution to the Facebook AI Image Similarity Challenge : Matching Track on DrivenData. The solution is based on self-supervised learning, and Vision Transformer(ViT). The main breaktrough comes from concatenating query and reference image to form as one image and asking ViT to directly predict from the image if query image used reference image. The solution scored 0.8291 Micro-average Precision on the private leaderboard.