V-JEPA 2.0 g384 attentive probe, ssv2 video classification, converted for LibreYOLO.

Converted from facebook/vjepa2-vitg-fpc64-384-ssv2 at revision 9f5fd615cb6f79065a28edcf1cc3ef25010dddfa.

The upstream checkpoint is remapped into LibreYOLO's native V-JEPA 2 module and wrapped with LibreYOLO v1.0 checkpoint metadata (family, size, task, dataset variant, clip geometry, preprocessing and the pooling rule). Tensor values are unchanged: the conversion is a key remap, loaded strictly, and the self-supervised predictor tower is dropped as a named, asserted set rather than by substring match.

Parity against unmodified transformers==5.1.0 on float32 CPU is exact (max_abs_diff == 0.0) for single-view logits.

This probe applies the exact upstream three-layer attentive pooler and linear classifier. One temporal view is used by default; the published multi-view accuracy is NOT claimed from single-view inference.

V-JEPA 2 is trained on video. An image is accepted as a single-frame input, which is a static appearance representation, not a motion one.

These weights are MIT, inherited from the source checkpoint above. The full text is in LICENSE, with attribution in NOTICE. Licences in this family differ per artifact, so do not assume a family-wide licence.

The probe was trained on a third-party video dataset, named for provenance only. Its dataset terms are not the terms of these weights, and LibreYOLO does not mirror or auto-download it.