Research map / Video / Video segmentation and tracking
Estimating People Flows to Better Count Them in Crowded Scenes
Weizhe Liu, Mathieu Salzmann, Pascal Fua
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 |
|---|---|---|---|
| regress people flows with strong conservation constraints instead of densities (Ours combined loss vs CAN) | CAN backbone crowd counter · CrowdFlow | MAE: 124.3 -> 97.8 | Improved |
| regress people flows with strong conservation constraints instead of densities (Ours combined loss vs CAN) (ablation) | CAN backbone crowd counter · FDST | MAE: 2.44 -> 2.17 | Improved |
| regress people flows with strong conservation constraints instead of densities (Ours combined loss vs CAN) (ablation) | CAN backbone crowd counter · UCSD | MAE: 0.98 -> 0.86 | Improved |
| add temporal averaging of three density maps (ablation) | CAN backbone crowd counter · CrowdFlow | MAE: 124.3 -> 128.9 | Hurt |
| feed two consecutive images for direct density regression (ablation) | CAN backbone crowd counter · CrowdFlow | MAE: 124.3 -> 125.7 | Hurt |
| add weaker conservation loss term (ablation) | CAN backbone crowd counter · CrowdFlow | MAE: 124.3 -> 121.2 | Improved |
| replace weak constraints by strong flow-based constraints (ablation) | CAN backbone crowd counter · CrowdFlow | MAE: 121.2 -> 97.8 | Improved |
| add forward-backward constraints (ablation) | CAN backbone crowd counter · CrowdFlow | MAE: 113.3 -> 97.8 | Improved |
| add forward-backward constraints (ablation) | CAN backbone crowd counter · FDST | MAE: 2.31 -> 2.17 | Improved |
| add forward-backward constraints (ablation) | CAN backbone crowd counter · UCSD | MAE: 0.94 -> 0.86 | Improved |
| regress flows (no forward-backward) instead of densities from two images (ablation) | CAN backbone crowd counter · CrowdFlow | MAE: 125.7 -> 113.3 | Improved |
| add PWC-Net optical flow correlation to the loss | CAN backbone crowd counter · FDST | MAE: 2.17 -> 2.10 | Improved |
Closest papers by meaning
- Counting People by Estimating People Flows (TPAMI 2021 · similarity 0.96)
- Context-Aware Crowd Counting (CVPR 2019 · similarity 0.88)
- Crowd Counting With Deep Structured Scale Integration Network (ICCV 2019 · similarity 0.87)
- Leveraging Self-Supervision for Cross-Domain Crowd Counting (CVPR 2022 · similarity 0.86)
- Almost Unsupervised Learning for Dense Crowd Counting (AAAI 2019 · similarity 0.86)
- Shallow Feature Based Dense Attention Network for Crowd Counting (AAAI 2020 · 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.
