Research map / Detection and segmentation / CNN features
Locate, Size and Count: Accurately Resolving People in Dense Crowds via Detection
Deepak Babu Sam, Skand Vishwanath Peri, Mukuntha Narayanan Sundararaman, Amogh Kamath, R. Venkatesh Babu
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 4 scale branches instead of 1 (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 155.2 -> 66.4 | Improved |
| add a fifth scale branch (nS=5 vs 4) (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 66.4 -> 69.3 | Hurt |
| use 3 boxes per scale instead of 1 (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 104.7 -> 66.4 | Improved |
| use 4 boxes per scale instead of 3 (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 66.4 -> 74.3 | Hurt |
| add top-down feature modulation (multiple connections) (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 94.5 -> 66.4 | Improved |
| use connections from all lower scales instead of sequential only (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 73.4 -> 66.4 | Improved |
| concatenate top-down features instead of multiplicative gating (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 67.6 -> 66.4 | No clear effect |
| add GWTA training (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 79.2 -> 66.4 | Improved |
| add class weighting of the loss (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 360.1 -> 66.4 | Improved |
| replicate VGG blocks in the feature extractor (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 79.3 -> 66.4 | Improved |
| classify discretized box sizes per pixel instead of regressing sizes (ablation) | LSC-CNN · ShanghaiTech Part_A | MAE: 77.9 -> 66.4 | Improved |
| detection output with LSC-CNN instead of thresholded density regression | LSC-CNN vs CSR-A-thr · ShanghaiTech Part_A | MLE (pixels): 16.8 -> 9.6 | Improved |
Closest papers by meaning
- Point in, Box Out: Beyond Counting Persons in Crowds (CVPR 2019 · similarity 0.91)
- Top-Down Feedback for Crowd Counting Convolutional Neural Network (AAAI 2018 · similarity 0.91)
- AutoScale: Learning to Scale for Crowd Counting (IJCV 2022 · similarity 0.91)
- Context-Aware Crowd Counting (CVPR 2019 · similarity 0.91)
- Divide and Grow: Capturing Huge Diversity in Crowd Images With Incrementally Growing CNN (CVPR 2018 · similarity 0.90)
- Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds (ECCV 2018 · 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.
