Research map / Image generation / Text-to-image diffusion
YOLO-Count: Differentiable Object Counting for Text-to-Image Generation
Guanning Zeng, Xiang Zhang, Zirui Wang, Haiyang Xu, Zeyuan Chen, Bingnan Li, Zhuowen Tu
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 LVIS strong-supervision pretraining (ablation) | YOLO-Count · FSC147 test | MAE: 18.42 -> 14.80 | Improved |
| add FSC147 weak-supervision finetuning (ablation) | YOLO-Count · FSC147 test | MAE: 43.91 -> 14.80 | Improved |
| use cardinality map regression instead of density map (ablation) | YOLO-Count · FSC147 test | MAE: 16.71 -> 14.80 | Improved |
| use cardinality map regression instead of density map (ablation) | YOLO-Count · FSC147 val | RMSE: 76.42 -> 58.36 | Improved |
| add classification branch for representation alignment (ablation) | YOLO-Count · FSC147 test | MAE: 17.01 -> 14.80 | Improved |
| add separate VLPAN feature output for each head (ablation) | YOLO-Count · FSC147 test | MAE: 16.54 -> 14.80 | Improved |
| use cardinality map instead of density map training (ablation) | regression counting models · FSC147 object-size downscaling | count bias | Improved |
| filter cardinality output with classification probability threshold 0.5 instead of 0.0 (ablation) | YOLO-Count · FSC147 | MAE | Hurt |
| filter cardinality output with classification probability threshold 0.5 instead of 0.0 | YOLO-Count · LVIS | MAE | Improved |
| use YOLO-Count guidance instead of CountGD surrogate loss or CLIP-Count | SDXL-Turbo token optimization · LargeGen and LargeGen-New | count accuracy | Improved |
| use YOLO-Count instead of CountGD | counting model · LVIS val | MAE: 4.84 -> 1.65 | Improved |
Closest papers by meaning
- Make It Count: Text-to-Image Generation with an Accurate Number of Objects (CVPR 2025 · similarity 0.89)
- CountGD: Multi-Modal Open-World Counting (NeurIPS 2024 · similarity 0.89)
- T2ICount: Enhancing Cross-modal Understanding for Zero-Shot Counting (CVPR 2025 · similarity 0.87)
- VLCounter: Text-Aware Visual Representation for Zero-Shot Object Counting (AAAI 2024 · similarity 0.87)
- OmniCount: Multi-label Object Counting with Semantic-Geometric Priors (AAAI 2025 · similarity 0.86)
- Learning to Count without Annotations (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.
