Research map / 3D vision / Human pose estimation
MAPConNet: Self-supervised 3D Pose Transfer with Mesh and Point Contrastive Learning
Jiaze Sun, Zhixiang Chen, Tae-Kyun Kim
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 mesh-level and point-level contrastive losses to supervised 3D-CoreNet | 3D-CoreNet · SMPL (supervised) | PMD: 0.36 -> 0.30 | Improved |
| add latent disentanglement (separate pose and identity channels) to unsupervised 3D-CoreNet (ablation) | 3D-CoreNet unsupervised · SMPL (unsupervised) | PMD: 14.09 -> 0.76 | Improved |
| add mesh and point contrastive losses on top of latent disentanglement (unsupervised) (ablation) | 3D-CoreNet unsupervised · DFAUST (unsupervised) | PMD: 92.45 -> 61.22 | Improved |
| add point-level contrastive loss (lambda_p=1) in supervised setting (ablation) | MAPConNet · SMPL (supervised) | PMD: 0.36 -> 0.35 | Improved |
| add mesh-level contrastive loss on SC (lambda_m,s=1) in supervised setting (ablation) | MAPConNet · SMPL (supervised) | PMD: 0.36 -> 0.31 | Improved |
| add both mesh and point losses in supervised setting (ablation) | MAPConNet · SMPL (supervised) | PMD: 0.36 -> 0.30 | Improved |
| add point loss in unsupervised setting (ablation) | MAPConNet · SMPL (unsupervised) | PMD: 0.76 -> 0.59 | Improved |
| add SC mesh loss on top of CC mesh loss (unsupervised) (ablation) | MAPConNet · SMPL (unsupervised) | PMD: 0.60 -> 0.59 | No clear effect |
| add unlabelled meshes to 50% labelled training (semi-supervised) | MAPConNet · SMPL | PMD: 3.69 -> 0.40 | Improved |
| add unlabelled DFAUST to SMPL-supervised training | MAPConNet · DFAUST | PMD: 120.10 -> 36.23 | Improved |
| use mesh loss without point loss (ablation) | MAPConNet · SMPL warped output | warped output shrinking | Mixed |
Later papers that compared with this method
1 later accepted paper reports MAPConNet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- DiLO: Disentangled Latent Optimization for Learning Shape and Deformation in Grouped Deforming 3D Objects (AAAI 2026 · better on 3 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
- Weakly-supervised 3D Pose Transfer with Keypoints (ICCV 2023 · similarity 0.90)
- Geometry-Contrastive Transformer for Generalized 3D Pose Transfer (AAAI 2022 · similarity 0.90)
- Unsupervised 3D Pose Transfer With Cross Consistency and Dual Reconstruction (TPAMI 2023 · similarity 0.89)
- Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose (NeurIPS 2021 · similarity 0.88)
- Diffusion-Driven Self-Supervised Learning for Shape Reconstruction and Pose Estimation (TPAMI 2025 · similarity 0.88)
- Towards Robust 3D Pose Transfer with Adversarial Learning (CVPR 2024 · similarity 0.87)
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.
