Research map / Detection and segmentation / Open-set and few-shot detection
Pixel-Wise Crowd Understanding via Synthetic Data
Qi Wang, Junyu Gao, Wenbin Lin, Yuan Yuan
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 |
|---|---|---|---|
| add segmentation as a multi-task objective (MTL vs single task) (ablation) | SFCN · GCC, random split | MAE: 36.2 -> 33.9 | Improved |
| add segmentation as a multi-task objective (MTL vs single task) (ablation) | SFCN · GCC, cross-camera split | MAE: 56.0 -> 52.6 | Improved |
| add segmentation as a multi-task objective (MTL vs single task) (ablation) | SFCN · GCC, cross-location split | MAE: 89.3 -> 85.7 | Improved |
| pre-train on GCC instead of ImageNet then fine-tune (ablation) | SFCN† · UCF-QNRF | MAE: 114.8 -> 102.0 | Improved |
| pre-train on GCC plus real datasets (Pre-GU) instead of ImageNet (ablation) | SFCN† · UCF-QNRF | MAE: 114.8 -> 98.6 | Improved |
| pre-train on GCC instead of training from scratch (ablation) | MCNN · UCF-QNRF | MAE: 281.2 -> 199.8 | Improved |
| pre-train on GCC instead of ImageNet, evaluate cross-dataset on SHT A (ablation) | SFCN† · SHT A | MAE: 108.0 -> 101.4 | Improved |
| pre-train on GCC instead of ImageNet, evaluate cross-dataset on SHT B (ablation) | SFCN† · SHT B | MAE: 17.2 -> 14.2 | Improved |
| use more GCC pre-training data (20% to 100%) (ablation) | SFCN · UCF-QNRF and SHT B | MAE and MSE | Improved |
| pre-train on high-luminance GCC data instead of low-luminance (ablation) | SFCN · UCF-QNRF | MAE: 126.7 -> 117.2 | Improved |
| pre-train on denser scene levels L5-L6 instead of L0-L2 (ablation) | SFCN · UCF-QNRF | MAE: 131.4 -> 117.0 | Improved |
| use ResNet-101 backbone (SFCN dagger) instead of VGG-16 SFCN (ablation) | SFCN · GCC, random split | MAE: 36.2 -> 28.1 | Improved |
Closest papers by meaning
- Learning From Synthetic Data for Crowd Counting in the Wild (CVPR 2019 · similarity 0.90)
- Leveraging Self-Supervision for Cross-Domain Crowd Counting (CVPR 2022 · similarity 0.89)
- Synthetic Data for Video Surveillance Applications of Computer Vision: A Review (IJCV 2024 · similarity 0.87)
- CrowdCLIP: Unsupervised Crowd Counting via Vision-Language Model (CVPR 2023 · similarity 0.87)
- MOTSynth: How Can Synthetic Data Help Pedestrian Detection and Tracking? (ICCV 2021 · similarity 0.87)
- Single Domain Generalization for Crowd Counting (CVPR 2024 · similarity 0.85)
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.
