DINOv3 (AI model): news timeline & CVEs

TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining

TT-VidT pairs a DINOv3 spatial ViT with a temporal layer, leading motion benchmarks at lower FLOPs.

TT-VidT separates a DINOv3-initialized ViT-B/16 spatial path from a compact Temporal Transfer Layer trained with Diff Compression to reconstruct target frames from a first-frame appearance anchor and motion tokens. A matched 24-configuration study at roughly 170–190M scale uses sim1.7M OpenVid and Moments-in-Time v2 clips for eight epochs. TT-VidT leads Jester, Something-Something V2, ARID, and Diving48, improving 54–121% over the strongest non-TT row while using 48% fewer encoder FLOPs than DisMo and 55% fewer than VideoMAE or V-JEPA2.

Hugging Face daily papersSep 26, 20:00 UTC · 6d agoAI research

Cross-modal learning for SAR target recognition using optical vision foundation models

Frozen DINOv3 optical prototypes supervise SAR target recognition without EO/SAR pairs, improving classification on the heavily imbalanced UNICORNv2 dataset.

The framework aligns SAR embeddings to class-level prototypes built from a frozen DINOv3 electro-optical encoder, requiring no strict EO/SAR image pairs. At inference the SAR model operates independently without access to optical imagery. On UNICORNv2, a civilian vehicle dataset with heavy speckle and severe class imbalance, EO prototype alignment improves accuracy over frozen DINOv3, SAR-only finetuning, and unpaired distribution alignment baselines.

arXiv cs.AI / cs.LG / cs.CLSep 7, 16:54 UTC · 25d agoAI research

DINOv3

Timeline2 articles

Appears with

TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining

Cross-modal learning for SAR target recognition using optical vision foundation models