Research map / Image generation / Autoregressive and latent generation
TaskTok: Delving into Task Tokens for Task-driven Image Restoration
Hongjae Lee, Sojung Kang, Jaeseong Yu, Seung-Won Jung
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 pre-restoration module before tokenizer (ablation) | TiTok-64 (Mix-B) · ImageNet classification | Top-1 accuracy (%): 47.4 -> 48.5 | Improved |
| add pre-restoration module before tokenizer (ablation) | TiTok-64 (Mix-B) · PASCAL VOC2012 segmentation | mIoU: 55.5 -> 56.9 | Improved |
| add pre-restoration module before tokenizer (ablation) | TiTok-64 (Mix-B) · PASCAL VOC2012 detection | mAP: 22.3 -> 23.1 | Improved |
| add token refiner updating all tokens (ablation) | TiTok-64 (Mix-B) with pre-restoration · ImageNet classification | Top-1 accuracy (%): 48.5 -> 46.3 | Hurt |
| add token refiner updating all tokens (ablation) | TiTok-64 (Mix-B) with pre-restoration · PASCAL VOC2012 detection | mAP: 23.1 -> 21.8 | Hurt |
| add token-level loss L_tok (ablation) | TiTok-64 with refiner · ImageNet classification | Top-1 accuracy (%): 46.3 -> 46.7 | No clear effect |
| add learnable token switch (ablation) | TiTok-64 with refiner and L_tok · ImageNet classification | Top-1 accuracy (%): 46.7 -> 46.9 | Improved |
| add mask regularizer L_mask (ablation) | TaskTok-64 with token switch · ImageNet classification | Top-1 accuracy (%): 46.9 -> 50.9 | Improved |
| add importance-order initialization pi (ablation) | TaskTok-64 with L_mask · ImageNet classification | Top-1 accuracy (%): 50.9 -> 53.0 | Improved |
| remove codebook quantization (continuous tokens) (ablation) | TaskTok-64 · ImageNet classification | Top-1 accuracy (%): 53.0 -> 55.3 | Improved |
| select tokens for another task (Seg/Det) when evaluating CLS (ablation) | TaskTok-64 · ImageNet classification | Top-1 accuracy (%): -1.4% / -1.8% from 55.3 | Hurt |
| select tokens across Seg and Det tasks (ablation) | TaskTok-64 · PASCAL VOC2012 | mIoU/mAP: -0.6% / -0.6% | Hurt |
Closest papers by meaning
- Debiased All-in-one Image Restoration with Task Uncertainty Regularization (AAAI 2025 · similarity 0.87)
- End-to-End Vision Tokenizer Tuning (NeurIPS 2025 · similarity 0.86)
- TransPrune: Token Transition Pruning for Efficient Large Vision-Language Model (CVPR 2026 · similarity 0.86)
- Controlling Vision-Language Models for Multi-Task Image Restoration (ICLR 2024 · similarity 0.86)
- Sharing Key Semantics in Transformer Makes Efficient Image Restoration (NeurIPS 2024 · similarity 0.86)
- LearnPruner: Rethinking Attention-based Token Pruning in Vision Language Models (ICLR 2026 · 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.
