Learning and Leveraging World Models in Visual Representation Learning
This work addresses the need for more flexible self-supervised visual representation learning, offering a method to control abstraction levels, but it is incremental as it builds on existing Joint-Embedding Predictive Architecture approaches.
The paper tackles the problem of generalizing self-supervised learning beyond masked image modeling by introducing Image World Models (IWM), which predict global photometric transformations in latent space, and shows that fine-tuned IWM matches or surpasses previous self-supervised methods in performance.
Joint-Embedding Predictive Architecture (JEPA) has emerged as a promising self-supervised approach that learns by leveraging a world model. While previously limited to predicting missing parts of an input, we explore how to generalize the JEPA prediction task to a broader set of corruptions. We introduce Image World Models, an approach that goes beyond masked image modeling and learns to predict the effect of global photometric transformations in latent space. We study the recipe of learning performant IWMs and show that it relies on three key aspects: conditioning, prediction difficulty, and capacity. Additionally, we show that the predictive world model learned by IWM can be adapted through finetuning to solve diverse tasks; a fine-tuned IWM world model matches or surpasses the performance of previous self-supervised methods. Finally, we show that learning with an IWM allows one to control the abstraction level of the learned representations, learning invariant representations such as contrastive methods, or equivariant representations such as masked image modelling.