Mask-supervised Object-centric Representation Learning with LeJEPA

Name: Mask-supervised Object-centric Representation Learning with LeJEPA
Uploaded: 2026-10-09T04:39:10.295999+00:00
Duration: 390 s
Description: Object-centric learning improves data efficiency by aligning representations at object level instead of scene level. Object LeJEPA outperforms image-level methods across tracking, classification, segmentation, and re-identification tasks on COCO data. It achieves strong performance even when trained on only 10% of the full dataset.

Object-centric learning improves data efficiency by aligning representations at object level instead of scene level. Object LeJEPA outperforms image-level methods across tracking, classification, segmentation, and re-identification tasks on COCO data. It achieves strong performance even when trained on only 10% of the full dataset.

Open the full episode