Research map / Label-efficient and robust learning / Self-supervised and contrastive
SeiT++: Masked Token Modeling Improves Storage-efficient Training
Minhyun Lee, Song Park, Byeongho Heo, Dongyoon Han, Hyunjung Shim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 masked token modeling pretraining on top of augmentations (ablation) | SeiT++ · ImageNet-1k (1.4 GB) | top-1 accuracy (%): 75.5 -> 77.8 | Improved |
| add masked token modeling to original SeiT (ablation) | SeiT · ImageNet-1k (1.4 GB) | top-1 accuracy (%): 74.0 -> 75.1 | Improved |
| add MTM at the smallest storage budget (ablation) | SeiT++ · ImageNet-1k (0.3 GB) | top-1 accuracy (%): 51.2 -> 60.6 | Improved |
| add ColorAdapt to SeiT (ablation) | SeiT · ImageNet-100 | top-1 accuracy (%): 77.3 -> 78.3 | Improved |
| add TokenAdapt to SeiT (ablation) | SeiT · ImageNet-100 | top-1 accuracy (%): 77.3 -> 80.4 | Improved |
| combine TokenAdapt and ColorAdapt (ablation) | SeiT · ImageNet-100 | top-1 accuracy (%): 77.3 -> 81.4 | Improved |
| use learned TokenAdapt module instead of decode-augment-reencode pipeline (ablation) | SeiT · ImageNet-100 (ViT-S) | top-1 accuracy (%): 80.1 -> 81.4 | Improved |
| reduce TokenAdapt training data to 100 images (ablation) | TokenAdapt · ImageNet-1k | top-1 accuracy (%): 75.5 -> 75.1 | No clear effect |
| apply token augmentations to robustness evaluation | SeiT · ImageNet-C Gaussian noise | top-1 accuracy (%): 50.7 -> 58.6 | Improved |
| apply SeiT++ to alternative tokenizers and DCT input (ablation) | SeiT · ImageNet-100 | top-1 accuracy (%): 81.8 -> 83.9 (VQGAN), 69.5 -> 71.0 (DCT) | Improved |
| reduce training images in data-scarce regime | SeiT++ · ImageNet-100 (13k images) | top-1 accuracy (%): 43.2 -> 52.3 | Improved |
Closest papers by meaning
- On the Role of Discrete Tokenization in Visual Representation Learning (ICLR 2024 · similarity 0.86)
- VQToken: Neural Discrete Token Representation Learning for Extreme Token Reduction in Video Large Language Models (NeurIPS 2025 · similarity 0.86)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.86)
- Vision-centric Token Compression in Large Language Model (NeurIPS 2025 · similarity 0.86)
- Token Cropr: Faster ViTs for Quite a Few Tasks (CVPR 2025 · similarity 0.85)
- Morphing Tokens Draw Strong Masked Image Models (ICLR 2025 · 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.
