1/4 ๐ข AHOY! accepted at #ECCV2026! ๐
From a single in-the-wild YouTube video, often heavily occluded, AHOY builds a complete, animatable 3D human avatar.
Each clip below runs input โ 360ยฐ avatar โ animation. ๐งต๐
๐ https://t.co/Uu7ksWqYzW
1/5 Dropping a 3D Gaussian Splatting avatar into a captured 3DGS scene is easy. Making it cast a believable shadow is not.
Happy to share RAGA, accepted at #ECCV2026.
4/4 The avatars are drivable and drop cleanly into 3DGS scenes captured on a phone. State of the art on occluded video.
๐ arXiv:2603.17975
๐ https://t.co/Uu7ksWqYzW
w/ @Xiangjun_Tang@peter_wonka@GerardPonsMoll1
3/4 The key idea: identity-finetuned video diffusion models hallucinate the unobserved regions and supervise the full body.
This turns partial, occluded views into a complete 3D Gaussian human that stays consistent from every angle.
Very grateful for the @eccvconf Program Chairs for all their efforts they spent on appeals! I think allowing appeals is very important for the community and provides valuable feedback to PCs and ACs. Something @CVPR and @ICCVConference have lost.
Very grateful for the @eccvconf Program Chairs for all their efforts they spent on appeals! I think allowing appeals is very important for the community and provides valuable feedback to PCs and ACs. Something @CVPR and @ICCVConference have lost.