Research map / Label-efficient and robust learning / Test-time adaptation
Continual Test-Time Adaptation for Single Image Defocus Deblurring via Causal Siamese Networks
Shuang Cui, Yi Li, Jiangmeng Li, Xiongxin Tang, Bing Su, Fanjiang Xu, Hui Xiong
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 CauSiam to source SIDD model | DPDNet-S · avg of 4 SIDD test sets | PSNR: 24.676 -> 25.412 | Improved |
| add CauSiam to source SIDD model | NRKNet · avg of 4 SIDD test sets | PSNR: 26.283 -> 26.560 | Improved |
| add CoTTA to source SIDD model | DPDNet-S · avg of 4 SIDD test sets | PSNR: 24.676 -> 22.863 | Hurt |
| add TENT-continual (entropy minimization) | DPDNet-S · avg of 4 SIDD test sets | PSNR: 24.676 -> 18.690 | Hurt |
| add CoTTA over three rounds | DRBNet · long-term CTTA | PSNR: 27.409 -> 23.933 | Hurt |
| add CauSiam over three rounds | DRBNet · long-term CTTA | PSNR: 27.409 -> 27.648 | Improved |
| add VLM-guided semantic priors (VSPI) (ablation) | CauSiam on DPDNet-S · avg of 4 SIDD test sets | PSNR: 25.077 -> 25.412 | Improved |
| add spatial consistency loss L_s (ablation) | CauSiam on DPDNet-S · avg of 4 SIDD test sets | PSNR: 24.910 -> 25.412 | Improved |
| add high-frequency consistency loss L_f (ablation) | CauSiam on DPDNet-S · avg of 4 SIDD test sets | PSNR: 25.384 -> 25.412 | Improved |
| update only the CA module instead of all online-model parameters (ablation) | CauSiam on DPDNet-S · avg of 4 SIDD test sets | PSNR: 25.146 -> 25.412 | Improved |
| add EMA update of offline model (ablation) | CauSiam on DPDNet-S · avg of 4 SIDD test sets | PSNR: 25.323 -> 25.412 | Improved |
| use DINOv2 instead of CLIP for priors (ablation) | CauSiam on DPDNet-S · avg of 4 SIDD test sets | PSNR: 25.412 -> 25.077 | Hurt |
Closest papers by meaning
- Single Image Defocus Deblurring via Implicit Neural Inverse Kernels (ICCV 2023 · similarity 0.88)
- Test-Time Fast Adaptation for Dynamic Scene Deblurring via Meta-Auxiliary Learning (CVPR 2021 · similarity 0.87)
- Reblurring-Guided Single Image Defocus Deblurring: A Learning Framework with Misaligned Training Pairs (IJCV 2025 · similarity 0.86)
- Neumann Network With Recursive Kernels for Single Image Defocus Deblurring (CVPR 2023 · similarity 0.86)
- Residual Diffusion Deblurring Model for Single Image Defocus Deblurring (AAAI 2025 · similarity 0.86)
- Prior and Prediction Inverse Kernel Transformer for Single Image Defocus Deblurring (AAAI 2024 · similarity 0.84)
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.
