Research map / Detection and segmentation / Weakly supervised detection and segmentation
Towards Partial Supervision for Generic Object Counting in Natural Scenes
Hisham Cholakkal, Guolei Sun, Salman Khan, Fahad Shahbaz Khan, Ling Shao, Luc Van Gool
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 dual-branch (classification + density) architecture instead of CAM+MSE regression (ablation) | LC framework · PASCAL VOC 2007 count-test | mRMSE: 0.45 -> 0.29 | Improved |
| add spatial loss to class + MSE loss (ablation) | LC framework · PASCAL VOC 2007 count-test | mRMSE-nz: 1.52 -> 1.32 | Improved |
| add ranking loss on top of class + spatial + MSE (ablation) | LC framework · PASCAL VOC 2007 count-test | mRMSE-nz: 1.32 -> 1.14 | Improved |
| set ranking loss weight lambda to 0.1 instead of 1 (ablation) | LC framework · PASCAL VOC 2007 count-test | mRMSE: 0.36 -> 0.29 | Improved |
| define lower-count range up to 4 (t=5) instead of up to 2 (t=3) (ablation) | LC framework · COCO count-test | mRMSE | Improved |
| use ground-truth point annotations for the spatial mask (ablation) | LC framework · PASCAL VOC 2007 count-test | mRMSE: 0.29 -> 0.28 | No clear effect |
| increase number of count-annotated categories from 20 to 80 (ablation) | RLC framework · COCO count-test | mRMSE: 0.41 -> 0.35 | Improved |
| use bottleneck hidden size P/2 instead of P in weight modulation layer (ablation) | RLC framework · COCO count-test | mRMSE-nz: 1.98 -> 1.94 | Improved |
| use 2 layers instead of 3 in weight modulation layer (ablation) | RLC framework · COCO count-test | mRMSE-nz: 2.07 -> 1.98 | Improved |
| use softmax instead of ReLU activation in weight modulation layer (ablation) | RLC framework · COCO count-test | mRMSE-nz: 1.97 -> 1.94 | Improved |
| block backpropagation from weight modulation layer to classifier (ablation) | RLC framework · COCO count-test (60/20 split) | mRMSE: 0.38 -> 0.36 | Improved |
| multiply category-specific density by total density map D_tot (ablation) | RLC framework · COCO count-test (60/20 split) | mRMSE: 0.37 -> 0.36 | Improved |
Closest papers by meaning
- Learning to Count without Annotations (CVPR 2024 · similarity 0.88)
- Point Segment and Count: A Generalized Framework for Object Counting (CVPR 2024 · similarity 0.86)
- OmniCount: Multi-label Object Counting with Semantic-Geometric Priors (AAAI 2025 · similarity 0.85)
- DAVE - A Detect-and-Verify Paradigm for Low-Shot Counting (CVPR 2024 · similarity 0.85)
- A Novel Unified Architecture for Low-Shot Counting by Detection and Segmentation (NeurIPS 2024 · similarity 0.85)
- Efficient Masked AutoEncoder for Video Object Counting and A Large-Scale Benchmark (ICLR 2025 · similarity 0.84)
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.
