Research map / Architectures and efficiency / Attention and transformers
Sapiens2
Rawal Khirodkar, He Wen, Julieta Martinez, Yuan Dong, Zhaoen Su, Shunsuke Saito
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 |
|---|---|---|---|
| scale Sapiens2 from 0.4B to 5B parameters | Sapiens2 · Pose 308 keypoints (11K test) | mAP (%): 76.9 -> 82.3 | Improved |
| scale Sapiens2 from 0.4B to 5B parameters | Sapiens2 · Albedo estimation (10K test) | MAE: 0.01825 -> 0.01191 | Improved |
| replace Sapiens-1B with Sapiens2-1B (in-the-wild supervision, 1K output) | Sapiens-1B · Body-part segmentation (5K test, 29 classes) | mIoU (%): 53.8 -> 81.7 | Improved |
| combine MAE with contrastive pretraining (Sapiens2-5B vs DINOv3-7B) | DINOv3-7B · Dense probing (frozen backbone) | pose mAP: 68.2 -> 74.7 | Improved |
| combine MAE with contrastive pretraining (Sapiens2-5B vs DINOv3-7B) | DINOv3-7B · Dense probing (frozen backbone) | albedo MAE (x1e-2): 3.48 -> 3.12 | Improved |
| use 4K hierarchical windowed backbone instead of 1K for 1B model | Sapiens2-1B · Body-part segmentation (5K test, 29 classes) | mIoU (%): 81.7 -> 81.9 | Improved |
| use 4K hierarchical windowed backbone instead of 1K for 1B model | Sapiens2-1B · Normal estimation (10K whole-body test) | mean angular error (deg): 7.12 -> 6.98 | Improved |
| use Sapiens2-5B instead of Sapiens2-1B | Sapiens2 · Body-part segmentation (5K test, 29 classes) | mAcc (%): 92.0 -> 91.1 | Hurt |
| use Sapiens2-1B instead of Sapiens2-0.8B | Sapiens2 · Pointmap (10K test) | RMSE: 0.459 -> 0.478 | Hurt |
Closest papers by meaning
- SapiensID: Foundation for Human Recognition (CVPR 2025 · similarity 0.89)
- HyperHuman: Hyper-Realistic Human Generation with Latent Structural Diffusion (ICLR 2024 · similarity 0.87)
- SapiensID 2.0: Aligning Human Recognition Foundation Models with Human Perception (NeurIPS 2026 · similarity 0.87)
- CapHuman: Capture Your Moments in Parallel Universes (CVPR 2024 · similarity 0.87)
- OpenHumanVid: A Large-Scale High-Quality Dataset for Enhancing Human-Centric Video Generation (CVPR 2025 · similarity 0.86)
- Not Too Generative, Not Too Discriminative: The Human Alignment Sweet Spot (NeurIPS 2026 · 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.
