Research map / Label-efficient and robust learning / Self-supervised and contrastive
Revisiting the Transferability of Supervised Pretraining: An MLP Perspective
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 |
|---|---|---|---|
| Add an MLP projector before the classifier | SL with ResNet50 · ImageNet-1K eval-D concept generalization, 100 epochs | top-1 accuracy: 55.9 -> 63.1 | Helped |
| Add an MLP projector before the classifier | SL with ResNet50 · ImageNet-1K eval-D concept generalization, 300 epochs | top-1 accuracy: 54.4 -> 64.1 | Helped |
| Add an MLP projector before the classifier | SL with ResNet34, ResNet101, Swin-tiny, and EfficientNetb2 · ImageNet-1K eval-D concept generalization | top-1 accuracy: 50.1 -> 55.0; 50.2 -> 55.8; 56.0 -> 63.6; 53.9 -> 64.7; 58.9 -> 60.6; 57.6 -> 64.2 | Helped |
| Add an MLP projector before the classifier | SL with ResNet50 · COCO object detection, 100 epochs pretraining | AP: 38.9 -> 39.7 | Helped |
| Add an MLP projector before the classifier | SL with ResNet50 · COCO object detection, 300 epochs pretraining | AP: 40.1 -> 40.7 | Helped |
| Add an MLP projector before the classifier | SL with ResNet50 · 12-domain classification datasets, linear evaluation average | top-1 accuracy: 75.89 -> 81.68 | Helped |
| Add an MLP projector before the classifier | SL with ResNet50 · 12-domain classification datasets, finetuning with 1000 training samples (average) | top-1 accuracy: 70.54 -> 71.66 | Helped |
| Add an MLP projector before the classifier | SL with ResNet50 · 12-domain classification datasets, 5-ways 5-shots few-shot (average) | top-1 accuracy: 67.76 -> 71.71 | Helped |
| Add an MLP projector (ablation) | Byol · ImageNet-1K concept generalization, layer4-pooled-features | top-1 accuracy: +23.3% | Helped |
| Add an MLP projector (ablation) | Mocov1 · ImageNet-1K concept generalization, layer4-pooled-features | top-1 accuracy: +5.1% | Helped |
| Add an MLP projector (ablation) | Mocov2 · ImageNet-1K concept generalization, layer4-pooled-features | top-1 accuracy: +3.7% | Helped |
| Use only an input fully connected layer (ablation) | SL with ResNet50 · ImageNet-1K eval-D concept generalization, 100 epochs | top-1 accuracy: 55.9 -> 56.6 | Helped |
Closest papers by meaning
- Boost Supervised Pretraining for Visual Transfer Learning: Implications of Self-Supervised Contrastive Representation Learning (AAAI 2022 · similarity 0.90)
- What Makes Instance Discrimination Good for Transfer Learning? (ICLR 2021 · similarity 0.90)
- On the Surprising Effectiveness of Attention Transfer for Vision Transformers (NeurIPS 2024 · similarity 0.87)
- Beyond Random Augmentations: Pretraining with Hard Views (ICLR 2025 · similarity 0.87)
- On the Provable Advantage of Unsupervised Pretraining (ICLR 2024 · similarity 0.87)
- Enhancing Pre-trained Representation Classifiability can Boost its Interpretability (ICLR 2025 · 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.