Research map / Detection and segmentation / CNN features
Does Robustness on ImageNet Transfer to Downstream Tasks?
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 |
|---|---|---|---|
| Use vanilla Swin-T instead of robustified CNN backbones | Full-network transfer to COCO detection and ADE20K segmentation · COCO and ADE20K under common corruptions | performance drop | Helped |
| Use ImageNet-robustified models rather than Regular | ResNet50 transfer to CIFAR10 · CIFAR10 classification under corruptions | performance drop | Hurt |
| Use an adversarially trained ImageNet model, especially the ε=0.1 model | Full-network transfer to COCO and ADE20K · COCO and ADE20K under common corruptions | performance drop | Hurt |
| Increase Swin model size while holding pretraining data at ImageNet-1k | Swin Transformer transfer · ADE20K and COCO under common corruptions | performance drop | Mixed |
| Pretrain the larger Swin Base model on ImageNet-22k instead of ImageNet-1k | Swin Base transfer · ADE20K semantic segmentation | IoU performance drop | Helped |
| Apply PRIME during transfer learning (ablation) | PRIME-pretrained ResNet50 · COCO object detection | Box AP clean: 36.42 -> 31.47 | Hurt |
| Apply PRIME during transfer learning (ablation) | PRIME-pretrained ResNet50 · COCO object detection | Box AP Drop: 22.29 % -> 21.21 % | Helped |
| Apply PRIME during transfer learning (ablation) | PRIME-pretrained ResNet50 · ADE20K segmentation | IoU clean: 40.64 -> 40.91 | Helped |
| Apply PRIME during transfer learning (ablation) | PRIME-pretrained ResNet50 · ADE20K segmentation | IoU corrupted: 28.56 -> 30.56 | Helped |
| Apply PRIME during transfer learning (ablation) | PRIME-pretrained ResNet50 · ADE20K segmentation | IoU Drop: 29.73 % -> 25.31 % | Helped |
| Use a smaller fine-tuning learning rate (ablation) | PRIME-pretrained ImageNet models · COCO and ADE20K transfer | percentage performance drop | Helped |
| Apply PRIME during transfer learning (ablation) | ImageNet-pretrained models · COCO object detection | clean performance | Hurt |
Closest papers by meaning
- Do Adversarially Robust ImageNet Models Transfer Better? (NeurIPS 2020 · similarity 0.90)
- ImageNet-E: Benchmarking Neural Network Robustness via Attribute Editing (CVPR 2023 · similarity 0.90)
- Vision Transformers Are Robust Learners (AAAI 2022 · similarity 0.88)
- ConvNet vs Transformer, Supervised vs CLIP: Beyond ImageNet Accuracy (ICML 2024 · similarity 0.88)
- Harnessing Edge Information for Improved Robustness in Vision Transformers (AAAI 2024 · similarity 0.87)
- 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.