Research map / 3D vision / Human pose estimation
CRAVES: Controlling Robotic Arm With a Vision-Based Economic System
Yiming Zuo, Weichao Qiu, Lingxi Xie, Fangwei Zhong, Yizhou Wang, Alan L. Yuille
Check 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 geometric constraints to refine keypoints and fine-tune on unlabeled real images | Synthetic-pretrained keypoint model · YouTube dataset | PCK@0.2 (%): 80.05 -> 87.01 | Improved |
| Use geometric constraints to refine keypoints and fine-tune on unlabeled real images | Synthetic-pretrained keypoint model · Lab dataset | PCK@0.2 (%): 95.66 -> 99.55 | Improved |
| Use geometric constraints to refine keypoints and fine-tune on unlabeled real images | Synthetic-pretrained pose estimator · Lab dataset | Average joint angular error (degrees): 7.13 -> 4.81 | Improved |
| Add background augmentation during fine-tuning (ablation) | Real-image keypoint model · Lab dataset | PCK@0.2 (%): 94.52 -> 98.72 | Improved |
| Add background augmentation during fine-tuning (ablation) | Real-image keypoint model · YouTube dataset | PCK@0.2 (%): 80.24 -> 84.04 | Improved |
| Add joint keypoint detection and 3D pose estimation (ablation) | Real-image keypoint model · YouTube dataset | PCK@0.2 (%): 80.24 -> 86.52 | Improved |
| Add joint keypoint detection and 3D pose estimation (ablation) | Real-image keypoint model · Lab dataset | PCK@0.2 (%): 94.52 -> 97.31 | Improved |
| Combine background augmentation with joint keypoint detection and 3D pose estimation (ablation) | Real-image keypoint model · YouTube dataset | PCK@0.2 (%): 80.24 -> 87.01 | Improved |
| Increase synthetic training images from 2,500 to 5,000 (ablation) | Synthetic-pretrained keypoint model · YouTube dataset | PCK@0.2 (%): 75.24 -> 80.05 | Improved |
| Increase synthetic training images from 5,000 to 20,000 (ablation) | Synthetic-pretrained keypoint model · YouTube dataset | PCK@0.2 (%): 80.05 -> 81.11 | Improved |
| Use the vision-based controller rather than human control from camera input | Reaching task · Reaching task | Success rate: 66.7% -> 59.3% | Hurt |
| Use the vision-based controller rather than human control from camera input | Reaching task · Reaching task | Average time (s): 38.8 -> 21.2 | Improved |
Closest papers by meaning
- Unsupervised Learning of Visual 3D Keypoints for Control (ICML 2021 · similarity 0.87)
- Pre-training Auto-regressive Robotic Models with 4D Representations (ICML 2025 · similarity 0.86)
- RoMan-4D: Learning Robot Arm Manipulation from 4D World Models (ECCV 2026 · similarity 0.86)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.85)
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.85)
- PointZero: 3D Point Track Completion for Learning Transferable 3D Dynamics (NeurIPS 2026 · similarity 0.85)
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.
