Research map / 3D vision / Face and shape
Unsupervised Volumetric Animation
Aliaksandr Siarohin, Willi Menapace, Ivan Skorokhodov, Kyle Olszewski, Jian Ren, Hsin-Ying Lee, Menglei Chai, Sergey Tulyakov
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 |
|---|---|---|---|
| use PnP-based pose prediction rather than directly predicting part poses (ablation) | UVA, geometry phase · VoxCeleb | AYD: 0.707 -> 0.141 | Improved |
| use PnP-based pose prediction rather than directly predicting part poses (ablation) | UVA, geometry phase · VoxCeleb | ASC: 0.160 -> 0.113 | Improved |
| add the unsupervised background loss (ablation) | UVA, geometry phase · VoxCeleb | AYD: 0.301 -> 0.141 | Improved |
| add geometry pretraining before introducing multiple parts (ablation) | UVA, 10 parts · VoxCeleb | AYD: 1.08 -> 0.051 | Improved |
| use five source frames instead of one | UVA · VoxCeleb | AYD: 0.051 -> 0.045 | Improved |
| use five source frames instead of one | UVA · VoxCeleb | L1: 0.0655 -> 0.0418 | Improved |
| use five source frames instead of one | UVA · TEDXPeople | AYD: 0.128 -> 0.107 | Improved |
| use five source frames instead of one | UVA · TEDXPeople | L1: 0.0474 -> 0.029 | Improved |
| use five source frames instead of one | UVA · VoxCeleb novel view synthesis | ASC: 0.119 -> 0.126 | Hurt |
| use five source frames instead of one | UVA · VoxCeleb novel view synthesis | APC: 0.171 -> 0.184 | Hurt |
| learn geometry and parts from monocular data without camera supervision | UVA · VoxCeleb | depth correlation: 0.53 -> 0.63 | Improved |
| learn geometry from real-image training data rather than use the general-purpose depth predictor | UVA · Synthetic faces | depth correlation: 0.602 -> 0.793 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Omnidata — Omnidata: A Scalable Pipeline for Making Multi-Task Mid-Level Vision Datasets From 3D Scans (ICCV 2021) · better on 2 of 2 shared results
- EpiGRAF — EpiGRAF: Rethinking training of 3D GANs (NeurIPS 2022) · better on 1 of 1 shared result
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
- BANMo: Building Animatable 3D Neural Models From Many Casual Videos (CVPR 2022 · similarity 0.88)
- Articulate your NeRF: Unsupervised articulated object modeling via conditional view synthesis (NeurIPS 2024 · similarity 0.88)
- LASR: Learning Articulated Shape Reconstruction From a Monocular Video (CVPR 2021 · similarity 0.88)
- Unsupervised Learning of Dense Shape Correspondence (CVPR 2019 · similarity 0.88)
- Unsupervised 3D Reconstruction and Grouping of Rigid and Non-Rigid Categories (TPAMI 2020 · similarity 0.87)
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · similarity 0.86)
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.
