Research map / 3D vision / Human pose estimation
Self-Supervised Learning of Interpretable Keypoints From Unlabelled Videos
Check 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 a conditional image generator to CycleGAN (ablation) | CycleGAN on Simplified Human3.6M · Simplified Human3.6M | landmark detection error (%): 3.54 -> 3.60 | Hurt |
| Add a skeleton bottleneck (ablation) | CycleGAN-based model · Simplified Human3.6M | landmark detection error (%): 3.54 -> 2.86 | Helped |
| Add a skeleton bottleneck (ablation) | CycleGAN · 300-W | landmark detection error (%): 11.89 -> 9.64 | Helped |
| Remove the second cycle constraint (ablation) | Conditional-generator and skeleton-bottleneck model · Simplified Human3.6M | landmark detection error (%): 2.86 -> 2.73 | Helped |
| Remove the second cycle constraint (ablation) | Conditional-generator and skeleton-bottleneck model · 300-W | landmark detection error (%): 9.64 -> 8.67 | Helped |
| Remove the second cycle constraint (ablation) | CycleGAN · Simplified Human3.6M | landmark detection error (%): 3.54 -> 3.39 | Helped |
| Reduce unpaired pose-prior samples from the full dataset to 5000 (ablation) | KeypointGAN · Simplified Human3.6M | keypoint localization error (%): 2.73 -> 2.92 ± 0.05 | Hurt |
| Reduce unpaired pose-prior samples from the full dataset to 500 (ablation) | KeypointGAN · 300-W | keypoint localization error (%): 8.67 -> 8.91 ± 0.15 | No clear effect |
| Reduce unpaired pose-prior samples from the full dataset to 50 (ablation) | KeypointGAN · 300-W with supervised post-processing | keypoint localization error (%): 5.12 -> 5.19 ± 0.06 | No clear effect |
| Use MPI-INF-3DHP rather than Human3.6M as the pose prior | KeypointGAN · Human3.6M | MSE (pixels): 14.46 -> 18.94 | Hurt |
| Add supervised post-processing | KeypointGAN · 300-W | keypoint localization error (%): 8.67 -> 5.12 | Helped |
Closest papers by meaning
- Self-Supervised 3D Human Pose Estimation via Part Guided Novel Image Synthesis (CVPR 2020 · similarity 0.88)
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · similarity 0.88)
- Learning Temporal Pose Estimation from Sparsely-Labeled Videos (NeurIPS 2019 · similarity 0.88)
- Learning Pose Specific Representations by Predicting Different Views (CVPR 2018 · similarity 0.87)
- Dual Pose-invariant Embeddings: Learning Category and Object-specific Discriminative Representations for Recognition and Retrieval (CVPR 2024 · similarity 0.87)
- Emergence of a Shared Canonical Object Frame from In-the-Wild Videos (NeurIPS 2026 · 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.