Text-Conditional JEPA for Learning Semantically Rich Visual Representations
For vision-language pretraining, TC-JEPA offers a new feature-prediction paradigm that surpasses contrastive methods on fine-grained tasks.
TC-JEPA uses image captions to reduce prediction uncertainty in masked feature prediction, improving downstream performance and training stability. It outperforms contrastive methods on tasks requiring fine-grained visual understanding.
Image-based Joint-Embedding Predictive Architecture (I-JEPA) offers a promising approach to visual self-supervised learning through masked feature prediction. However with the inherent visual uncertainty at masked positions, feature prediction remains challenging and may fail to learn semantic representations. In this work, we propose Text-Conditional JEPA (TC-JEPA) that uses image captions to reduce the prediction uncertainty. Specifically, we modulate the predicted patch features using a fine-grained text conditioner that computes sparse cross-attention over input text tokens. With such conditioning, patch features become predictable as a function of text, thus are more semantically meaningful. We show TC-JEPA improves downstream performance and training stability, with promising scaling properties. TC-JEPA also offers a new vision-language pretraining paradigm based on feature prediction only, outperforming contrastive methods on diverse tasks, especially those requiring fine-grained visual understanding and reasoning.