Research map / Label-efficient and robust learning / Recognition and long-tail
On Robustness and Transferability of Convolutional Neural Networks
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 |
|---|---|---|---|
| Increase pre-training data and training duration | ResNet-101x3 · OOD benchmarks | relative classification error: 20-30% reduction | Helped |
| Increase model size from ResNet-50 to ResNet-101x3 | ImageNet-21k-pretrained models · OOD benchmarks | classification error: 5--20% | Helped |
| Train longer while increasing model size on a subset of ImageNet-21k rather than using the full dataset | ResNet-101x3 · OOD benchmarks | classification error | Hurt |
| Use a larger test-time resolution | Image classifiers · ImageNet-A and ObjectNet | accuracy | Helped |
| Apply FixRes using test-time preprocessing and higher resolution | Image classifiers · ImageNet-A and ObjectNet | accuracy | Helped |
| Apply FixRes | Image classifiers · ImageNet | accuracy | No clear effect |
| Increase test-time resolution and/or apply higher-resolution FixRes | Small EfficientNet and ResNet models · ImageNet-Vid-Robust and YouTube-BB-Robust | accuracy | Helped |
| Increase test-time resolution and/or apply higher-resolution FixRes | Larger EfficientNet and ResNet models · ImageNet-Vid-Robust and YouTube-BB-Robust | accuracy | No clear effect |
| Increase pre-training dataset size | ResNet-50 and ResNet-101x3 · SI-Score size, location, and rotation subsets | uniformity of accuracy across factor values | Helped |
| Increase model size | ResNet-101x3 vs ResNet-50 · SI-Score size, location, and rotation subsets | robustness to factor variation | Helped |
| Use GroupNorm instead of BatchNorm | ImageNet-trained ResNet-50 · SI-Score smaller objects | accuracy | Helped |
| Use BatchNorm instead of GroupNorm | ImageNet-trained ResNet-50 · SI-Score larger objects | accuracy | Helped |
Closest papers by meaning
- Measuring Robustness to Natural Distribution Shifts in Image Classification (NeurIPS 2020 · similarity 0.90)
- Do Adversarially Robust ImageNet Models Transfer Better? (NeurIPS 2020 · similarity 0.90)
- Revealing the Dark Secrets of Extremely Large Kernel ConvNets on Robustness (ICML 2024 · similarity 0.87)
- DRoP: Distributionally Robust Data Pruning (ICLR 2025 · similarity 0.87)
- Gain from Neighbors: Boosting Model Robustness in the Wild via Adversarial Perturbations Toward Neighboring Classes (CVPR 2025 · similarity 0.86)
- ImageNet-D: Benchmarking Neural Network Robustness on Diffusion Synthetic Object (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.