Research map / 3D vision / Human pose estimation
Triangulation Residual Loss for Data-efficient 3D Pose Estimation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 TR loss to baseline triangulation (ablation) | HRnet multiview triangulation · Human3.6M Subject 1 - Subject 11 | MPJPE (mm): 172.96 -> 38.80 | Helped |
| add TR loss to baseline triangulation (ablation) | HRnet multiview triangulation · CalMS21-THM | MPJPE (mm): 27.34 -> 14.65 | Helped |
| add confidence head to TR (ablation) | TR model · Human3.6M Subject 1 - Subject 11 | MPJPE (mm): 38.80 -> 37.54 | Helped |
| add confidence head to TR (ablation) | TR model · CalMS21-THM | MPJPE (mm): 14.65 -> 14.14 | Helped |
| add domain discriminator loss (ablation) | TR+confidence · CalMS21-THM | MPJPE (mm): 14.14 -> 14.05 | Helped |
| add domain discriminator loss (ablation) | TR+confidence · Human3.6M Subject 1 - Subject 11 | MPJPE (mm): 37.54 -> 37.20 | Helped |
| use 3D supervision instead of TR loss (ablation) | baseline triangulation · Human3.6M Subject 1 - Subject 11 | MPJPE (mm): 172.96 -> 175.36 | No clear effect |
| train with 2D labels plus TR loss instead of 3D supervised methods | TR loss (100%) · Human3.6M Protocol 1 | MPJPE (mm): 27.5 -> 25.8 | Helped |
| reduce 2D labeled training data from 100% to 5% | TR loss · Human3.6M Protocol 1 | MPJPE (mm): 25.8 -> 28.7 | Hurt |
| add TR loss to cross-dataset training instead of 3D-supervised finetuning | Sup3d* · Dannce-THM | MPJPE (mm): 7.33 -> 5.87 | Helped |
| add TR loss on top of 3D supervision | Sup3d* · Calms2-THM | MPJPE (mm): 14.46 -> 14.38 | Helped |
| add domain discrimination loss to TR in cross-dataset tasks (ablation) | TR · Dannce-THM | MPJPE (mm): 5.87 -> 5.78 | Helped |
Closest papers by meaning
- Lightweight Multi-View 3D Pose Estimation Through Camera-Disentangled Representation (CVPR 2020 · similarity 0.86)
- Geometry-Driven Self-Supervised Method for 3D Human Pose Estimation (AAAI 2020 · similarity 0.86)
- Multiview-Consistent Semi-Supervised Learning for 3D Human Pose Estimation (CVPR 2020 · similarity 0.85)
- BEAST3D: animal behavioral analysis and neural encoding from multi-view video via Gaussian splatting (NeurIPS 2026 · similarity 0.85)
- TokenHMR: Advancing Human Mesh Recovery with a Tokenized Pose Representation (CVPR 2024 · similarity 0.84)
- Not All Tasks Quantize Equally: Fisher-Guided Quantization for Visual Geometry Transformer (NeurIPS 2026 · 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.