Research map / Detection and segmentation / CNN features
Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds
Haroon Idrees, Muhmmad Tayyab, Kishan Athrey, Dong Zhang, Somaya Al-Maadeed, Nasir Rajpoot, Mubarak Shah
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 |
|---|---|---|---|
| connect Density Network to DenseBlock2 rather than DenseBlock4 (ablation) | Composition Loss network · UCF-QNRF count | C-MAE: 148 -> 132 | Improved |
| connect Density Network to DenseBlock2 rather than DenseBlock3 (ablation) | Composition Loss network · UCF-QNRF count | C-MAE: 144 -> 132 | Improved |
| use all three maps (add D_inf localization map) instead of D1 and D2 only (ablation) | Composition Loss network · UCF-QNRF count | C-MAE: 137 -> 132 | Improved |
| use D1 and D2 rather than D1 only (ablation) | Composition Loss network · UCF-QNRF count | C-MAE: 141 -> 137 | Improved |
| add multiple density levels with Composition Loss to the base DenseNet201 counter (ablation) | DenseNet201 · UCF-QNRF count | C-MAE: 163 -> 132 | Improved |
| keep density maps unconnected to the count output instead of concatenating map counts (ablation) | Composition Loss network · UCF-QNRF count | C-MAE: 139 -> 132 | Improved |
| keep density maps unconnected to the count output instead of taking mean of four counts (ablation) | Composition Loss network · UCF-QNRF count | C-MAE: 150 -> 132 | Improved |
| enforce mean of counts from all sources (Mean) instead of unconnected maps (ablation) | Composition Loss network · UCF-QNRF D_1 density map | MAE: 1629 -> 1151 | Mixed |
| replace best prior counting method (Densenet201) with proposed | Densenet201 · UCF-QNRF | C-MAE: 163 -> 132 | Improved |
| replace best prior localizer (Encoder-Decoder) with proposed | Encoder-Decoder · UCF-QNRF localization | L-AUC: 0.670 -> 0.714 | Improved |
Closest papers by meaning
- A Generalized Loss Function for Crowd Counting and Localization (CVPR 2021 · similarity 0.92)
- AutoScale: Learning to Scale for Crowd Counting (IJCV 2022 · similarity 0.91)
- Iterative Crowd Counting (ECCV 2018 · similarity 0.91)
- Learning to Count via Unbalanced Optimal Transport (AAAI 2021 · similarity 0.90)
- Crowd Counting and Density Estimation by Trellis Encoder-Decoder Networks (CVPR 2019 · similarity 0.90)
- Adaptive Density Map Generation for Crowd Counting (ICCV 2019 · similarity 0.90)
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.
