Research map / 3D vision / Human pose estimation
Rethinking Pose Estimation in Crowds: Overcoming the Detection Information Bottleneck and Ambiguity
Mu Zhou, Lucas Stoffl, Mackenzie Weygandt Mathis, Alexander Mathis
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 bottom-up detections plus pose proposals instead of an object detector with plain top-down | BUCTD · CrowdPose test | AP: 70.0 -> 78.5 | Improved |
| add CTD conditional top-down refinement to HigherHRNet-W32 detections | HigherHRNet-W32 · CrowdPose test | AP: 65.7 -> 71.4 | Improved |
| use CoAM conditioning instead of preNet (ablation) | CTD · CrowdPose test | AP: 69.5 -> 71.4 | Improved |
| inject conditions earlier in the network (stage 1 vs stage 4) (ablation) | BUCTD-CoAM-W32 · CrowdPose test | AP: 68.3 -> 71.2 | Improved |
| train with conditions from 15 checkpoints rather than one (ablation) | CTD-CoAM-W32 · CrowdPose test | AP: 68.8 -> 71.4 | Improved |
| use generative sampling for training conditions (ablation) | CTD-CoAM-W32 · CrowdPose test | AP: 71.4 -> 72.3 | Improved |
| use generative sampling on fish data (ablation) | BUCTD · SchoolingFish | AP: 88.7 -> 81.9 | Hurt |
| use bottom-up detections instead of object-detector boxes (ablation) | CTD · Marmosets hard frames | AP_H: 52.4 -> 89.4 | Improved |
| iteratively refine by feeding predictions back as conditions (3x) (ablation) | BUCTD · OCHuman test | AP: 47.4 -> 48.5 | Improved |
| apply BUCTD refinement on COCO with few overlapping people | PETR · COCO val | AP: 73.1 -> 77.8 | Improved |
Closest papers by meaning
- Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation (AAAI 2020 · similarity 0.90)
- DeeperCut: A Deeper, Stronger, and Faster Multi-person Pose Estimation Model (ECCV 2016 · similarity 0.89)
- CrowdPose: Efficient Crowded Scenes Pose Estimation and a New Benchmark (CVPR 2019 · similarity 0.89)
- Robust Pose Estimation in Crowded Scenes with Direct Pose-Level Inference (NeurIPS 2021 · similarity 0.88)
- Combining Detection and Tracking for Human Pose Estimation in Videos (CVPR 2020 · similarity 0.88)
- PoseTrack: A Benchmark for Human Pose Estimation and Tracking (CVPR 2018 · similarity 0.88)
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.
