Research map / Video / Video segmentation and tracking
Counting People by Estimating People Flows
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 between locations instead of densities (ablation) | CAN backbone · CrowdFlow | MAE: 124.3 -> 113.3 | Improved |
| replace flow regression with direct density regression from two images (ablation) | Ours · CrowdFlow | MAE: 97.8 -> 125.7 | Hurt |
| average densities over three images instead of using flows (ablation) | CAN · CrowdFlow | MAE: 124.3 -> 128.9 | Hurt |
| add weaker conservation loss on densities (ablation) | CAN · CrowdFlow | MAE: 124.3 -> 121.2 | Improved |
| add forward-backward cycle consistency constraints (ablation) | Flow regressor · CrowdFlow | MAE: 113.3 -> 97.8 | Improved |
| add optical flow regularization from PWC-Net (ablation) | Ours (combined loss) · CrowdFlow | MAE: 97.8 -> 96.3 | Improved |
| use ground-truth instead of estimated optical flow | Ours · CrowdFlow | MAE: 96.3 -> 90.9 | Improved |
| feed original images to the optical flow regressor in addition to density maps (ablation) | Ours (PWC-Net optical flow) · CrowdFlow | MAE: 96.3 -> 97.5 | Hurt |
| add spatial loss term to the fully supervised loss (ablation) | Ours (combined loss) · CrowdFlow | MAE: 97.8 -> 97.7 | No clear effect |
| average forward and backward flows rather than using one (ablation) | Ours · UCSD | MAE: 0.87 -> 0.86 | Improved |
| work in the ground plane instead of the image plane | Ours (combined loss) · Venice | MAE: 15.0 -> 12.3 | Improved |
| annotate every 2nd frame instead of every frame (ablation) | Ours (combined loss) · CrowdFlow | MAE: 97.8 -> 98.1 | No clear effect |
Closest papers by meaning
- Estimating People Flows to Better Count Them in Crowded Scenes (ECCV 2020 · similarity 0.96)
- Context-Aware Crowd Counting (CVPR 2019 · similarity 0.91)
- Point in, Box Out: Beyond Counting Persons in Crowds (CVPR 2019 · similarity 0.89)
- Divide and Grow: Capturing Huge Diversity in Crowd Images With Incrementally Growing CNN (CVPR 2018 · similarity 0.89)
- Top-Down Feedback for Crowd Counting Convolutional Neural Network (AAAI 2018 · similarity 0.89)
- Almost Unsupervised Learning for Dense Crowd Counting (AAAI 2019 · similarity 0.89)
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.
