Dividing and Conquering Cross-Modal Recipe Retrieval: from Nearest Neighbours Baselines to SoTA
This work addresses the problem of cross-modal retrieval for recipes and other domains, offering a simpler and more effective approach, though it is incremental as it builds on existing encoders and methods.
The authors tackled cross-modal recipe retrieval by proposing a non-parametric method that, when combined with standard encoders, outperforms most existing methods and, with triplet loss, improves state-of-the-art on the Recipe1M dataset by a large margin, while also generalizing to other domains like Politics and GoodNews.
We propose a novel non-parametric method for cross-modal recipe retrieval which is applied on top of precomputed image and text embeddings. By combining our method with standard approaches for building image and text encoders, trained independently with a self-supervised classification objective, we create a baseline model which outperforms most existing methods on a challenging image-to-recipe task. We also use our method for comparing image and text encoders trained using different modern approaches, thus addressing the issues hindering the development of novel methods for cross-modal recipe retrieval. We demonstrate how to use the insights from model comparison and extend our baseline model with standard triplet loss that improves state-of-the-art on the Recipe1M dataset by a large margin, while using only precomputed features and with much less complexity than existing methods. Further, our approach readily generalizes beyond recipe retrieval to other challenging domains, achieving state-of-the-art performance on Politics and GoodNews cross-modal retrieval tasks.