Research map / Vision-language and multimodal / Human and embodied modeling
Compositional Attribute Imbalance in Vision Datasets
Yanbiao Ma, Jiayi Chen, Wei Dai, Dong Zhao, Zeyu Zhang, Yuting Yang, Bowei Liu, Jiaxuan Zhao, Andi Zhang
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 CAS-based sampling weights to CutMix | CutMix/FMix/SaliencyMix with ResNeXt50 · ImageNet-1k | top-1 accuracy (%): 49.78 -> 50.96 | Improved |
| add CAS-based sampling weights to SaliencyMix | CutMix/FMix/SaliencyMix with ResNeXt50 · ImageNet-1k | top-1 accuracy (%): 47.35 -> 50.42 | Improved |
| add CAS-based sampling weights to CutMix | CutMix/FMix/SaliencyMix with ResNeXt50 · ImageNet-1k (high-CAS subset) | top-1 accuracy (%): 30.25 -> 33.79 | Improved |
| add CAS-based sampling weights to CutMix | CutMix/FMix/SaliencyMix with ResNet18 · Stanford Cars | top-1 accuracy (%): 82.49 -> 89.57 | Improved |
| add CAS-based sampling weights to FMix | CutMix/FMix/SaliencyMix with ResNet18 · Stanford Cars | top-1 accuracy (%): 75.72 -> 86.51 | Improved |
| add CAS-based sampling weights to CutMix | CutMix/FMix/SaliencyMix with ResNeXt50 · Stanford Dogs | top-1 accuracy (%): 69.14 -> 74.57 | Improved |
| add CAS-based sampling weights to CutMix | CutMix/FMix/SaliencyMix with ResNet18 · CIFAR-100 | top-1 accuracy (%): 54.57 -> 55.28 | Improved |
| add CAS-based sampling weights (all augmentations) | CutMix/FMix/SaliencyMix with ResNeXt50 · DTD | top-1 accuracy (%): +0.09 for FMix (99.56 -> 99.65) | No clear effect |
| add CAS-sampling to CutMix on CLIP backbone vs DINOv2 | CutMix/FMix/SaliencyMix with DINOv2 · DTD | overall gain: +1.38 (CLIP) vs +0.97 (DINOv2) | Mixed |
| set power parameter b to 1.2 (ablation) | CutMix/FMix/SaliencyMix with ResNeXt50 · CIFAR-100 and ImageNet | top-1 accuracy (%) | Improved |
| apply CAS sampling on top of CutMix | CutMix/FMix/SaliencyMix with ResNet18 · ImageNet-1K | mean CAS std: 36.8 -> 29.3 | Improved |
Closest papers by meaning
- Deep Imbalanced Attribute Classification using Visual Attention Aggregation (ECCV 2018 · similarity 0.90)
- Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent (NeurIPS 2025 · similarity 0.86)
- Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning (AAAI 2024 · similarity 0.86)
- Compositional Caching for Training-free Open-vocabulary Attribute Detection (CVPR 2025 · similarity 0.86)
- Scalable Evaluation and Neural Models for Compositional Generalization (NeurIPS 2025 · similarity 0.86)
- Does Data Scaling Lead to Visual Compositional Generalization? (ICML 2025 · 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.
