Google SIGGRAPH paper rebuilds 3D heads in 0.08 seconds with 88% less GPU memory
rsasaki0109 · x · 2026-07-27
Google’s SIGGRAPH 2026 paper, “Topologically Consistent Multi-view 3D Head Reconstruction via Coarse-Guided Layered Surface Sampling,” introduces SHELLS, a method that reconstructs dense 3D heads from calibrated multi-view images.
Key results in the post:
- reconstructs 18k-vertex 3D heads in 0.08 seconds
- uses DINOv2 features aggregated through projective surface-aware feature sampling
- lets a transformer predict dense semantic meshes
- is 3.5× faster than state of the art
- uses 88% less GPU memory
The main claim is that this produces an efficient reconstruction pipeline with better global consistency across views.
More from Multimodal
- Screenshot any character, one prompt recreates it — iamaliveix · 2026-09-23
- Third-party test: Claude Opus 5.5 renders finer 3D scenes but costs 13x more than GPT-6 Sol — testingcatalog · 2026-09-23
- ComfyUI trick: aux preprocessor + Qwen transfers poses across characters with one prompt — Acceptable-Work8202 · 2026-09-23
- Same portrait prompt across Midjourney V6.1, V7 and V8.2: do older models look better? — tisch_eins · 2026-09-23
- Testing AI character consistency across a 20-image travel sequence — SiennaVaire · 2026-09-23
- Midjourney v8.2 Faces: New Portrait Generation Samples Shared — azed_ai · 2026-09-23