Research map / Architectures and efficiency / Attention and transformers
ImageNet-21K Pretraining for the Masses
Read the paperCheck 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 semantic softmax instead of single-label pretraining | ImageNet-21K-P pretraining · ImageNet-1K | top-1 accuracy (%): 81.1 -> 81.4 | Helped |
| Use multi-label rather than single-label pretraining | ImageNet-21K-P pretraining · ImageNet-1K | top-1 accuracy (%): 81.1 -> 81.0 | Hurt |
| Add semantic KD to semantic softmax pretraining (ablation) | ImageNet-21K-P pretraining · ImageNet-1K | top-1 accuracy (%): 81.4 -> 82.2 | Helped |
| Add semantic KD to semantic softmax pretraining (ablation) | ImageNet-21K-P pretraining · iNaturalist | top-1 accuracy (%): 72.0 -> 72.7 | Helped |
| Add semantic KD to semantic softmax pretraining (ablation) | ImageNet-21K-P pretraining · Food 251 | top-1 accuracy (%): 75.8 -> 76.1 | Helped |
| Add semantic KD to semantic softmax pretraining (ablation) | ImageNet-21K-P pretraining · CIFAR-100 | top-1 accuracy (%): 90.4 -> 91.7 | Helped |
| Add semantic KD to semantic softmax pretraining (ablation) | ImageNet-21K-P pretraining · MS-COCO | mAP (%): 81.3 -> 82.2 | Helped |
| Add semantic KD to semantic softmax pretraining (ablation) | ImageNet-21K-P pretraining · Pascal-VOC | mAP (%): 89.7 -> 89.8 | Helped |
| Add semantic KD to semantic softmax pretraining (ablation) | ImageNet-21K-P pretraining · Kinetics 200 | top-1 accuracy (%): 83.0 -> 84.4 | Helped |
| Use ASL rather than cross-entropy for multi-label pretraining (ablation) | ImageNet-21K-P pretraining · ImageNet-21K-P | macro-mAP (%): 73.9 -> 74.7 | Helped |
| Use ASL rather than cross-entropy for multi-label pretraining (ablation) | ImageNet-21K-P pretraining · ImageNet-1K | top-1 accuracy (%): 79.6 -> 81.0 | Helped |
| Use semantic softmax rather than official ImageNet-21K pretraining | ViT-B-16 · ImageNet-1K | top-1 accuracy (%): 83.3 -> 83.9 | Helped |
Closest papers by meaning
- Rethinking Supervised Pre-Training for Better Downstream Transferring (ICLR 2022 · similarity 0.87)
- Beyond Random Augmentations: Pretraining with Hard Views (ICLR 2025 · similarity 0.87)
- SimVLM: Simple Visual Language Model Pretraining with Weak Supervision (ICLR 2022 · similarity 0.87)
- ParameterNet: Parameters Are All You Need for Large-scale Visual Pretraining of Mobile Networks (CVPR 2024 · similarity 0.86)
- Quick-Tune: Quickly Learning Which Pretrained Model to Finetune and How (ICLR 2024 · similarity 0.86)
- Scalable Pre-training of Large Autoregressive Image Models (ICML 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.