- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
ESCD queries a frozen derainer under two aligned views and reuses its clean predictions and rain residuals for student distillation, conditioned rain synthesis, and spatial loss weighting. Subsequent cycles promote the student to the next frozen teacher while retaining the collaborative system. Only the student is required at inference.
Critical strengths
-
The framework connects restoration targets, synthesis conditioning, and spatial regularization through a shared teacher signal, without requiring an external semantic or quality-assessment model.
-
The method performs strongly across synthetic, real daytime, and nighttime benchmarks. Importantly, the complete Stage-II graph already improves Rain100L from 33.28 to 34.84 dB before iterative refinement.
-
Single-view, fusion, generator, weighting, and cycle-count comparisons support several design choices. The teacher, generator, and discriminator introduce no additional inference modules.
Critical weaknesses
-
The extra 0.68 dB on Rain100L follows further training cycles, without equal-budget continued-training or fixed-teacher controls. The acknowledged training-cost limitation therefore remains experimentally unresolved.
-
Teacher residuals can contain background textures and restoration errors, which subsequently influence all three paths. Quantitative analysis against available synthetic rain layers and tests with weaker teachers are needed to characterize error propagation.
-
The concatenation comparison supports SAFRG overall, but does not separate the effects of factorization, affine modulation, gating, and capacity. Separate encoders and identity cases alone do not demonstrate clean-content and rain disentanglement.
-
RGSUD is absent from the main unpaired-real perceptual comparisons, and the checkpoint-selection criterion and validation split are unspecified. Small perceptual-score differences also require uncertainty estimates.
Questions for authors
-
Do ISCD gains persist against continued training with a fixed teacher and other self-distillation controls under matched total computation?
-
How accurately do the residual priors identify rain, and how sensitive is the system to teacher errors or background-texture contamination?
-
Can targeted SAFRG ablations isolate factorization from modulation, gating, and additional parameters?
-
Please clarify checkpoint selection, any use of paired ground truth during selection, and comparisons with RGSUD under a common real-rain transfer protocol.