Research map / 3D vision / Face and shape
SHeaP: Self-Supervised Head Geometry Predictor Learned via 2D Gaussians
Liam Schoneveld, Zhe Chen, Davide Davoli, Jiapeng Tang, Saimon Terazawa, Ko Nishino, Matthias Nießner
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add perceptual loss (ablation) | photometric-only SHeaP · NoW validation | median distance (mm): 1.09 -> 1.06 | Improved |
| add identity perceptual loss (ablation) | SHeaP with perceptual loss · NoW validation | median distance (mm): 1.06 -> 0.99 | Improved |
| add normals+depth coupling loss (ablation) | SHeaP without densify/prune · NoW validation | median distance (mm): 0.99 -> 0.96 | Improved |
| add normals+depth coupling loss (ablation) | full SHeaP minus coupling · NoW validation | median distance (mm): 0.98 -> 0.93 | Improved |
| add densification and pruning (ablation) | SHeaP with ID loss and coupling · NoW validation | median distance (mm): 0.96 -> 0.93 | Improved |
| enable graph convolution edges (ablation) | SHeaP with MLP ResNet · NoW validation | median distance (mm): 0.98 -> 0.93 | Improved |
| add densify/prune and normals+depth together (ablation) | SHeaP with perceptual loss only · NoW validation | median distance (mm): 1.06 -> 1.03 | Improved |
| add ID loss to perceptual+densify+coupling (ablation) | SHeaP · NoW validation | median distance (mm): 1.03 -> 0.93 | Improved |
| add ID loss (ablation) | SHeaP with perceptual and densify/coupling · NoW validation | mean distance (mm): 1.25 -> 1.16 | Improved |
| replace mesh rendering with 2D Gaussians (ablation) | DECA-style mesh-rendering baseline · NoW validation | median distance (mm): 1.18 -> 1.09 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EMOCA — EMOCA: Emotion Driven Monocular Face Capture and Animation (CVPR 2022) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Gaussian Eigen Models for Human Heads (CVPR 2025 · similarity 0.87)
- Learning Complete 3D Morphable Face Models From Images and Videos (CVPR 2021 · similarity 0.86)
- Generating 3D Faces using Convolutional Mesh Autoencoders (ECCV 2018 · similarity 0.86)
- 3D Gaussian Head Avatars with Expressive Dynamic Appearances by Compact Tensorial Representations (CVPR 2025 · similarity 0.85)
- Gaussian Head & Shoulders: High Fidelity Neural Upper Body Avatars with Anchor Gaussian Guided Texture Warping (ICLR 2025 · similarity 0.84)
- Gaussian Head Avatar: Ultra High-fidelity Head Avatar via Dynamic Gaussians (CVPR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
