We argue that there are many notions of ‘similarity’ and that models, like humans, should be able to adapt to these dynamically. This contrasts with most representation learning methods, supervised or self-supervised, which learn a fixed embedding function and hence implicitly assume a single notion of similarity. For instance, models trained on ImageNet are biased towards object categories, while a user might prefer the model to focus on colors, textures or specific elements in the scene. In this paper, we propose the GeneCIS (‘genesis’) benchmark, which measures models’ ability to adapt to a range of similarity conditions. Extending prior work, our benchmark is designed for zeroshot evaluation only, and hence considers an open-set of similarity conditions. We find that baselines from powerful CLIP models struggle on GeneCIS and that performance on the benchmark is only weakly correlated with ImageNet accuracy, suggesting that simply scaling existing methods is not fruitful. We further propose a simple, scalable solution based on automatically mining information from existing image-caption datasets. We find our method offers a substantial boost over the baselines on GeneCIS, and further improves zero-shot performance on related image retrieval benchmarks. In fact, though evaluated zero-shot, our model surpasses state-of-the-art supervised models on MIT-States.

Our benchmark contains four conditional retrieval tasks. Given a reference image (yellow boxes) and text condition (blue text), the model must select the only conditionally similar image (green boxes) from a curated set of gallery images. The galleries contain hard distractor images (which are similar but not conditionally similar) to the reference image, to prevent shortcut solutions.

@inproceedings{vaze2023gen,
        title={GeneCIS: A Benchmark for General Conditional Image Similarity},
        author={Sagar Vaze and Nicolas Carion and Ishan Misra},
        booktitle={CVPR},
        year={2023}
        }