Research map / Image generation / Image restoration
TAPE: Task-Agnostic Prior Embedding for Image Restoration
Lin Liu, Lingxi Xie, Xiaopeng Zhang, Shanxin Yuan, Xiangyu Chen, Wengang Zhou, Houqiang Li, Qi Tian
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 task-agnostic pre-training (ablation) | TAPE-Net · Rain200L | PSNR: 31.72 -> 33.17 | Improved |
| add task-agnostic pre-training (ablation) | TAPE-Net · SIDD | PSNR: 37.41 -> 37.90 | Improved |
| add task-agnostic pre-training then fine-tune on unseen task (ablation) | TAPE-Net · Snow100K | PSNR: 25.42 -> 26.33 | Improved |
| add task-agnostic pre-training then fine-tune on unseen task (ablation) | TAPE-Net · ISTD | PSNR: 26.28 -> 26.57 | Improved |
| add pre-training for IPT versus TAPE on unseen tasks | IPT vs TAPE-Net · Snow100K/ISTD/REDS | PSNR gain: IPT boosts 0.07dB, 0.19dB and 0.08dB; TAPE improves 0.91dB, 0.29dB and 0.48dB | Improved |
| widen synthetic Gaussian noise range in pre-training from [5,20] to [1,50] (ablation) | TAPE-Net · SIDD | PSNR: 37.60 -> 37.72 | Improved |
| pre-train with 5 datasets instead of 3 (ablation) | TAPE-Net · Snow100K | PSNR: 26.17 -> 26.33 | Improved |
| add more tasks to pre-training (ablation) | TAPE-Net · pre-training-known tasks (Raindrop800, Rain200L, TIP2018) | PSNR: maximum PSNR difference is 0.04 dB | No clear effect |
| add pre-training of network theta (ablation) | TAPE-Swin · Rain200H | PSNR: 25.96 -> 26.18 | Improved |
| add pre-training of network phi (ablation) | TAPE-Swin · Rain200H | PSNR: 26.15 -> 26.18 | Improved |
| add pre-training of theta and PLM (ablation) | TAPE-Swin · Rain200H | PSNR: 25.73 -> 26.18 | Improved |
| add pixel-wise contrastive loss in pre-training (ablation) | TAPE-Net · Rain200L | PSNR: decrease by 0.12dB/0.001 without the loss | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning from History: Task-agnostic Model Contrastive Learning for Image Restoration (AAAI 2024 · similarity 0.87)
- Boosting Image Restoration via Priors from Pre-trained Models (CVPR 2024 · similarity 0.86)
- Debiased All-in-one Image Restoration with Task Uncertainty Regularization (AAAI 2025 · similarity 0.86)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.86)
- Controlling Vision-Language Models for Multi-Task Image Restoration (ICLR 2024 · similarity 0.85)
- Parameter Efficient Adaptation for Image Restoration with Heterogeneous Mixture-of-Experts (NeurIPS 2024 · 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.
