Tsinghua team turns two images into 3D mirror-illusion art with AutoMIA
新智元 · wechat · 2026-07-24
A Tsinghua-led team introduced AutoMIA, a system that automatically designs mirror-illusion 3D art from two target images.
- Users specify one image for the object’s front view and another for its mirror view.
- The method jointly optimizes 3D shape and color with differentiable rendering, and the resulting model can be 3D printed.
- The paper reports improvements over ShadowArt-style baselines, including better smoothness, lower noise, and higher shape consistency, while running efficiently on a single RTX 3090.
More from Embodied
- Independent study says Waymo vehicles had 68% fewer crashes per mile than human drivers — reed · 2026-07-24
- Mistral’s 8B Robostral Navigate uses one RGB camera and tops R2R-CE at 77.4% — mistralai · 2026-07-24
- REK teases a samurai robot warm-up before tonight’s Tokyo fight night — cixliv · 2026-07-24
- SAGE generates simulation-ready 3D scenes and releases a 10k embodied-AI dataset — rsasaki0109 · 2026-07-24
- Black Forest Labs says FLUX 3 beats major multimodal rivals and powers robotics — Latent Space · 2026-07-24
- ReferTrack tracks language-specified targets with one camera and reaches 89.4% on EVT-Bench — tencent · 2026-07-24