Research map / Label-efficient and robust learning / Recognition and long-tail
Representation Learning by Learning to Count
Mehdi Noroozi, Hamed Pirsiavash, Paolo Favaro
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 randomized mixture of downsampling methods during training (ablation) | Counting · PASCAL VOC 2007 detection | mAP: at least 2.2% | Improved |
| add mixed interpolation instead of Lanczos only (ablation) | Counting · PASCAL VOC 2007 detection | mAP: 47.3 -> 51.4 | Improved |
| add mixed interpolation instead of linear only (ablation) | Counting · PASCAL VOC 2007 detection | mAP: 48.4 -> 51.4 | Improved |
| mixed downsampling in training reduces sensitivity of pretext error to downsampling style at test time (ablation) | Counting · Downsampling-style pairwise error | std of error across test methods: 0.18 -> 0.04 | Improved |
| increase training set from 128K to 1.3M images (ablation) | Counting · PASCAL VOC 2007 detection | mAP: 44.9 -> 51.4 | Improved |
| increase training set from 512K to 1.3M images (ablation) | Counting · PASCAL VOC 2007 detection | mAP: 49.1 -> 51.4 | Improved |
| train with RGB/gray mixture instead of RGB only (ablation) | Counting · PASCAL VOC 2007 detection | mAP: 48.2 -> 51.4 | Improved |
| train with RGB/gray mixture instead of gray only (ablation) | Counting · PASCAL VOC 2007 detection | mAP: 50.4 -> 51.4 | Improved |
| use 1000-d counting vector instead of 20-d (ablation) | Counting · PASCAL VOC 2007 detection | mAP: 50.9 -> 51.4 | No clear effect |
Closest papers by meaning
- Trading robust representations for sample complexity through self-supervised visual experience (NeurIPS 2018 · similarity 0.86)
- Counting in Small Transformers: The Delicate Interplay between Attention and Feed-Forward Layers (ICML 2025 · similarity 0.84)
- Learning to Count without Annotations (CVPR 2024 · similarity 0.84)
- Towards flexible perception with visual memory (ICML 2025 · similarity 0.83)
- Vision Transformer Off-the-Shelf: A Surprising Baseline for Few-Shot Class-Agnostic Counting (AAAI 2024 · similarity 0.83)
- I-Con: A Unifying Framework for Representation Learning (ICLR 2025 · similarity 0.83)
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.
