3rd Place Solution to Large-scale Fine-grained Food Recognition

TL;DR

This work tackles large-scale, fine-grained food recognition on Food1K by building an ensemble of multiple backbones trained with a novel loss mix. The authors show that ArcFace alone performs strongly and that incorporating Circle loss with a carefully chosen weighting improves convergence and accuracy, formalized as $L = \gamma_{0}L_{a} + \gamma_{1}L_{c}$ with $\gamma_{0}=1$ and $\gamma_{1}=1/\beta$, where $\beta$ is the batch size. They further boost performance via distillation (KD with temperature $T=3$), test-time augmentation, and model ensemble, ultimately attaining 3rd place in the competition. The approach emphasizes diverse backbones, tuned augmentations, and weighted fusion of embeddings, highlighting the practical value of combining discriminative loss functions and cross-model collaboration for large-scale fine-grained recognition tasks.

Abstract

Food analysis is becoming a hot topic in health area, in which fine-grained food recognition task plays an important role. In this paper, we describe the details of our solution to the LargeFineFoodAI-ICCV Workshop-Recognition challenge held on Kaggle. We find a proper combination of Arcface loss[1] and Circle loss[9] can bring improvement to the performance. With Arcface and the combined loss, model was trained with carefully tuned configurations and ensembled to get the final results. Our solution won the 3rd place in the competition.