Research map / 3D vision / Human pose estimation
FeatER: An Efficient Network for Human Reconstruction via Feature Map-Based TransformER
Ce Zheng, Matias Mendieta, Taojiannan Yang, Guo-Jun Qi, Chen Chen
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 feature-map-preserving FeatER instead of the small vanilla transformer (ablation) | FeatER 2D pose pipeline · COCO val2017 | AP: 74.0 -> 74.9 | Improved |
| Replace FeatER with VanillaTransformer_L (ablation) | FeatER 2D pose pipeline · COCO val2017 | AP: 74.9 -> 75.1 | Mixed |
| Remove transformer refinement (ablation) | FeatER 2D pose pipeline · COCO val2017 | AP: 72.9 -> 74.9 | Hurt |
| Use FeatER instead of VanillaTransformer_S (ablation) | FeatER 3D pose and mesh pipeline · Human3.6M test | MPJPE: 52.7 -> 49.9 | Improved |
| Use FeatER instead of VanillaTransformer_L (ablation) | FeatER 3D pose and mesh pipeline · Human3.6M test | MPJPE: 48.3 -> 49.9 | Mixed |
| Remove transformer refinement (ablation) | FeatER 3D pose and mesh pipeline · Human3.6M test | MPJPE: 54.4 -> 49.9 | Hurt |
| Add feature map reconstruction to the 3D part (ablation) | FeatER 3D pose and mesh pipeline · Human3.6M test | MPJPE: 53.3 -> 49.9 | Improved |
| Add feature map reconstruction to the 3D part (ablation) | FeatER 3D pose and mesh pipeline · 3DPW test | MPVE: 106.9 -> 105.6 | Improved |
| Place feature map reconstruction only in the 2D part rather than the 3D part (ablation) | FeatER pose and mesh pipeline · Human3.6M test | MPJPE: 52.8 -> 49.9 | Hurt |
| Use reconstruction modules in both 2D and 3D parts rather than only in 3D (ablation) | FeatER pose and mesh pipeline · 3DPW test | MPVE: 86.9 -> 87.1 | Hurt |
| Use FeatER rather than MeshGraphormer | 3D pose reconstruction · Human3.6M test | MPJPE: 51.2 -> 49.9 | Improved |
| Use FeatER rather than the best listed baseline | 3D pose and mesh reconstruction · 3DPW test | PA-MPJPE: 45.2 -> 45.9 | Hurt |
Closest papers by meaning
- THUNDR: Transformer-Based 3D Human Reconstruction With Markers (ICCV 2021 · similarity 0.84)
- ART: Articulated Reconstruction Transformer (CVPR 2026 · similarity 0.83)
- TranSG: Transformer-Based Skeleton Graph Prototype Contrastive Learning With Structure-Trajectory Prompted Reconstruction for Person Re-Identification (CVPR 2023 · similarity 0.83)
- Joint Reconstruction of 3D Human and Object via Contact-Based Refinement Transformer (CVPR 2024 · similarity 0.81)
- GeneMAN: Generalizable Single-Image 3D Human Reconstruction from Multi-Source Human Data (NeurIPS 2025 · similarity 0.81)
- MagicMan: Generative Novel View Synthesis of Humans with 3D-Aware Diffusion and Iterative Refinement (AAAI 2025 · similarity 0.80)
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.
