Reproduce “LEF-JEPA: Denoising-Trained Latent Energy Fields for Spatial Representation Learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhiyuan Liu, Zhihao Xu, Sicheng Liu, Huanhuan Huang

LEF-JEPA trains visual encoders to represent object structure by learning a conditional energy field that guides perturbed patch features toward target features, then transferring this preference through a relative-energy loss. In frozen-encoder tests on CIFAR-10, PartImageNet, and Two-Room planning, its variants outperform V-JEPA2.1 across tasks and backbones; ViT-B CIM improves accuracy, foreground IoU, and success, within these evaluations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.