Research map / Detection and segmentation / Open-set and few-shot detection
ABC Easy as 123: A Blind Counter for Exemplar-Free Multi-Class Class-agnostic Counting
Michael A Hobley, Victor Adrian Prisacariu
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 exemplar-free blind multi-class counter ABC123 instead of exemplar-based LOCA | ABC123 · MCAC test set | MAE: 10.91 -> 9.52 | Improved |
| train the full ViT backbone rather than freezing it (ablation) | ABC123 · MCAC test set | MAE: 15.76 -> 9.52 | Improved |
| combine sub-class density maps by sum instead of max (ablation) | ABC123 · FSC-147 test (<300 objects) | MAE: 22.43 -> 11.75 | Improved |
| increase number of prediction heads from 5 to 100 (ablation) | ABC123 · MCAC test set | MAE: 9.52 -> 7.43 | Improved |
| increase number of prediction heads to 100 (ablation) | ABC123 · MCAC test set | head utilisation: 100% -> 39% | Hurt |
| use a single head with no matching stage (ablation) | ABC123 · MCAC-M1 test set (trained on MCAC-M1) | MAE: 7.54 -> 7.53 | No clear effect |
| train ABC123 on single-class MCAC-M1 instead of full MCAC (ablation) | ABC123 · MCAC-M1 test set | MAE: 6.82 -> 7.54 | No clear effect |
| train exemplar-based LOCA/BMNet on single-class MCAC-M1 instead of multi-class MCAC (ablation) | BMNet+ · MCAC-M1 test set | MAE: 11.57 -> 8.05 | Improved |
| apply ABC123 trained on MCAC to FSC-147 vs exemplar-based CounTR | ABC123 · FSC-147 test (<300 objects) | MAE: 21.09 -> 11.75 | Improved |
Closest papers by meaning
- Vision Transformer Off-the-Shelf: A Surprising Baseline for Few-Shot Class-Agnostic Counting (AAAI 2024 · similarity 0.89)
- A Fixed-Point Approach to Unified Prompt-Based Counting (AAAI 2024 · similarity 0.89)
- PBECount: Prompt-Before-Extract Paradigm for Class-Agnostic Counting (AAAI 2025 · similarity 0.88)
- Learning to Count without Annotations (CVPR 2024 · similarity 0.86)
- Referring Expression Counting (CVPR 2024 · similarity 0.85)
- Point Segment and Count: A Generalized Framework for Object Counting (CVPR 2024 · 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.
