Research map / Detection and segmentation / Open-set and few-shot detection
A Comprehensive Study of Image Classification Model Sensitivity to Foregrounds, Backgrounds, and Visual Attributes
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train ResNets with adversarial training rather than standard supervision | ResNets · RIVAL10, foreground and background noise | relative foreground sensitivity (RFS) | Hurt |
| Use contrastive training rather than standard supervised training | ResNets and ViTs · RIVAL10, foreground and background noise | relative foreground sensitivity (RFS) | Hurt |
| Increase corruption level | Transformers · RIVAL10, foreground and background noise | relative foreground sensitivity (RFS) | Helped |
| Increase corruption level | ResNets · RIVAL10, foreground and background noise | relative foreground sensitivity (RFS) | No clear effect |
| Use DeiT training instead of ViT training | Transformers · RIVAL10, foreground and background noise | relative foreground sensitivity (RFS) | Hurt |
| Increase ViT patch size from 16×16 to 32×32 (ablation) | ViT · RIVAL10, foreground and background noise | relative foreground sensitivity (RFS) | Hurt |
| Increase adversarial-training attack size ε (ablation) | Robust ResNets · RIVAL10, foreground and background noise | relative foreground sensitivity (RFS) | No clear effect |
| Use adversarial training | ResNets · RIVAL10 images with grayed backgrounds | accuracy | Hurt |
| Use transformer features rather than ResNet or SimCLR features | Linear classifier on frozen features · RIVAL10 images with grayed backgrounds | accuracy | Helped |
| Remove an individual attribute by graying its region (ablation) | Image classification models · RIVAL10 | accuracy: average drop of roughly 3.5%; most attribute-model pairs drop less than 1% | Hurt |
| Use saliency alignment alone to assess foreground sensitivity | GradCAM analysis · RIVAL10 | alignment with foreground | No clear effect |
| Use larger foreground sensitivity as a universal class property | Image classification models · RIVAL10 classes | relative foreground sensitivity (RFS) | Mixed |
Closest papers by meaning
- ImageNet-E: Benchmarking Neural Network Robustness via Attribute Editing (CVPR 2023 · similarity 0.88)
- Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation (CVPR 2025 · similarity 0.87)
- ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression (NeurIPS 2025 · similarity 0.86)
- Adaptive Sensitivity Analysis for Robust Augmentation against Natural Corruptions in Image Segmentation (ICML 2025 · similarity 0.85)
- Benchmarking the Attribution Quality of Vision Models (NeurIPS 2024 · similarity 0.85)
- ImageNet-D: Benchmarking Neural Network Robustness on Diffusion Synthetic Object (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.