A JEPA novelty score that appears to triage driving clips across datasets is actually detecting dataset provenance, and fails on a same-dataset benchmark where a supervised probe on the same embeddings succeeds.
In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Asleep at the Wheel: JEPA's Limitations in Evaluating Novel Driving Data
A JEPA novelty score that appears to triage driving clips across datasets is actually detecting dataset provenance, and fails on a same-dataset benchmark where a supervised probe on the same embeddings succeeds.