Research map / Vision-language and multimodal / CLIP and zero-shot
Boosting Quantitive and Spatial Awareness for Zero-Shot Object Counting
Da Zhang, Bingyu Li, Feiyu Wang, Zhiyuan Zhao, Junyu Gao
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 text prompts (no quantity) to CLIP fine-tuning baseline (ablation) | QICA (ViT-L/14) · FSC-147 test | MAE: 16.15 -> 14.89 | Improved |
| add synergistic quantity-aware prompting (SPS) (ablation) | QICA (ViT-L/14) · FSC-147 test | MAE: 14.89 -> 13.72 | Improved |
| add cost aggregation decoder (CAD) (ablation) | QICA (ViT-L/14) · FSC-147 test | MAE: 13.72 -> 12.88 | Improved |
| add multi-level quantity alignment loss (LMQA) (ablation) | QICA (ViT-L/14) · FSC-147 test | MAE: 12.88 -> 12.41 | Improved |
| use K=5 quantity hypotheses instead of K=1 (ablation) | QICA (ViT-L/14) · FSC-147 val | MAE: 13.76 -> 12.98 | Improved |
| increase hypotheses from K=5 to K=7 (ablation) | QICA (ViT-L/14) · FSC-147 val | MAE: 12.98 -> 13.12 | Hurt |
| joint SPS prompting in vision and language instead of vision-only prompting (ablation) | QICA (ViT-L/14) · FSC-147 val | MAE: 14.38 -> 12.98 | Improved |
| joint SPS prompting instead of language-only prompting (ablation) | QICA (ViT-L/14) · FSC-147 val | MAE: 14.52 -> 12.98 | Improved |
| use prompt length 16 instead of 2 (ablation) | QICA (ViT-L/14) · FSC-147 val | MAE: 12.98 -> 13.52 | Hurt |
| QICA ViT-L/14 vs T2ICount on FSC-147 test | QICA · FSC-147 test | MAE: 11.76 -> 12.41 | Hurt |
| QICA ViT-L/14 vs CountGD text mode on ShanghaiTech-A | QICA · ShanghaiTech-A | MAE: 141.9 -> 140.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- T2ICount — T2ICount: Enhancing Cross-modal Understanding for Zero-Shot Counting (CVPR 2025) · better on 2 of 3 shared results
- VLCounter — VLCounter: Text-Aware Visual Representation for Zero-Shot Object Counting (AAAI 2024) · better on 2 of 2 shared results
- CountGD* — CountGD: Multi-Modal Open-World Counting (NeurIPS 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VLCounter: Text-Aware Visual Representation for Zero-Shot Object Counting (AAAI 2024 · similarity 0.92)
- T2ICount: Enhancing Cross-modal Understanding for Zero-Shot Counting (CVPR 2025 · similarity 0.90)
- CountGD: Multi-Modal Open-World Counting (NeurIPS 2024 · similarity 0.87)
- Vision Transformer Off-the-Shelf: A Surprising Baseline for Few-Shot Class-Agnostic Counting (AAAI 2024 · similarity 0.87)
- Single Domain Generalization for Few-Shot Counting via Universal Representation Matching (CVPR 2025 · similarity 0.87)
- Point Segment and Count: A Generalized Framework for Object Counting (CVPR 2024 · similarity 0.86)
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.
