DeepGHS Teases AnimeDINOv3: 2B Anime-Focused DINOv3 Beats Legacy Features Across the Board
bdsqlsz · x · 2026-10-21
DeepGHS announced AnimeDINOv3, a 2B-parameter DINOv3 pretrained specifically on anime imagery, now in release preparation (HF repo already created, still private). Internal frozen-feature evaluations show gains across anime downstream tasks: recognition benchmarks (Tag probe AP 28.5, Zerochan kNN/linear, CSIP) match or beat the previous RG-2M 512 checkpoint and crush legacy Anime DINOv2 (CAFormer-b36), with similar sweeps on dense tasks (Hard3 parts AP 93.9, G3 seg IoU 95.0, G4 PCK@16, G5 parts).
More from Multimodal
- MiniMax-H3 hailed as favorite open release of the year, runs fully local — AdinaYakup · 2026-10-23
- Michal Malewicz experiments with depth maps in his creative workflow — michalmalewicz · 2026-10-23
- One-Shot Tunable Style Transfer on Flux3 Via Reference Latent + Node — Winter_unmuted · 2026-10-23
- PhysVidBench accepted at EMNLP 2026: do video gen models grasp physical commonsense? — aykuterdemml · 2026-10-23
- EMNLP 2026 paper teaches multimodal LLMs to get cartoons via incongruity-resolution supervision — aykuterdemml · 2026-10-23
- A gorgeous Midjourney style: sref code 820221253 for dreamy lavender watercolor looks — umesh_ai · 2026-10-23