QualiTeacher: Quality-Conditioned Pseudo-Labeling for Real-World Image Restoration
Authors: Fengyang Xiao, Jingjia Feng, Peng Hu, Dingming Zhang, Lei Xu, Guanyi Qin, Lu Li, Chunming He, Sina Farsiu
Organizations: Duke University · Tsinghua University · École Polytechnique Fédérale de Lausanne (EPFL) · National University of Singapore · Sun Yat-sen University
Real-world image restoration (RWIR) is a highly challenging task due to the absence of clean ground-truth images. Many recent methods resort to pseudo-label (PL) supervision, often within a Mean-Teacher (MT) framework. However, these methods face a critical paradox: unconditionally trusting the often imperfect, low-quality PLs forces the student model to learn undesirable artifacts, while discarding them severely limits data diversity and impairs model generalization. In this paper, we propose QualiTeacher, a novel framework that transforms pseudo-label quality from a noisy liability into a conditional supervisory signal. Instead of filtering, QualiTeacher explicitly conditions the student model on the quality of the PLs, estimated by an ensemble of complementary non-reference image quality assessment (NR-IQA) models spanning low-level distortion and semantic-level assessment. This strategy teaches the student network to learn a quality-graded restoration manifold, enabling it to understand what constitutes different quality levels. Consequently, it can not only avoid mimicking artifacts from low-quality labels but also extrapolate to generate results of higher quality than the teacher itself. To ensure the robustness and accuracy of this quality-driven learning, we further enhance the process with a multi-augmentation scheme to diversify the PL quality spectrum, a score-based preference optimization strategy inspired by Direct Preference Optimization (DPO) to enforce a monotonically ordered quality separation, and a cropped consistency loss to prevent adversarial over-optimization (reward hacking) of the IQA models. Experiments on standard RWIR benchmarks demonstrate that QualiTeacher can serve as a plug-and-play strategy to improve the quality of the existing pseudo-labeling framework, establishing a new paradigm for learning from imperfect supervision. Code will be released.
Figures & tables
Figure 1: Comparison of pseudo-label (PL) utilization strategies in the mean-teacher framework. (a) Unconditional Trust: PLs are used without filtering, causing degradation artifacts to be replicated by the student. (b) Aggressive Filtering: Low-quality PLs are discarded via NR-IQA filtering, yet over-smoothed outputs that receive deceptively high scores survive, introducing blurriness. (c) QualiTeacher (Ours): Nearly all PLs are retained (with only extreme outliers discarded), with their quality scores injected as continuous conditioning signals, enabling the student to leverage full data diversity while remaining aware of each sample’s reliability.
Figure 2: Overall framework of QualiTeacher. By re-purposing pseudo-label quality as a conditional signal rather than a filtering criterion, the student network can learn a quality-graded manifold and extrapolate beyond the teacher’s capabilities.
Figure 3: Overview of the quality-driven optimization strategy in QualiTeacher. Left: Cropped Quality Consistency Loss. Right: Score-Based Preference Optimization.
Figure 4: Visualizations on desnowing, deraining, dehazing, LLIE, and UIE.
Desnowing
Methods
CNNIQA ↑
HyperIQA ↑
ManIQA ↑
Topiq ↑
Arniqa ↑
Q-Align ↑
C2score ↑
LIQE ↑
SemiDDM-Weather [ 30 ]
0.636
0.423
0.342
0.442
0.567
3.586
55.068
2.497
SnowMaster (NAFNet) [ 19 ]
0.670
0.501
0.349
0.499
0.612
3.836
55.327
2.768
NAFNet + Colabator [ 10 ]
0.628
0.470
0.342
0.473
0.620
3.936
55.389
2.667
NAFNet + QualiTeacher
0.675
0.559
0.362
0.540
0.628
4.030
55.446
2.883
HistoFormer [ 34 ]
0.615
0.444
0.346
0.455
0.585
3.827
55.266
2.610
Table 1: NR-IQA comparisons on Desnowing, Deraining, Dehazing, LLIE, and UIE.
Table 6Figure 7
Methods
CNNIQA ↑
ManIQA ↑
Topiq ↑
Arniqa ↑
Q-Align ↑
C2score ↑
LIQE ↑
MFRQA ↑
CIDNet [ 46 ]
0.554
0.354
0.480
0.618
3.715
54.944
3.049
58.860
CIDNet + Colabator [ 10 ]
0.532
0.367
0.439
0.609
3.571
54.967
2.929
59.319
CIDNet + QualiTeacher1
0.660
0.432
0.530
0.661
3.643
55.454
3.411
59.675
CIDNet + QualiTeacher(Ours)
0.669
0.428
0.542
0.657
3.741
55.400
3.322
59.341
Table 4: Ablation on the IQA selection. QualiTeacher with alternative NR-IQA metrics (MANIQA, TOPIQ, QualiClip) on LLIE.
Methods (AP)
Bicycle
Boat
Bottle
Bus
Car
Cat
Chair
Cup
Dog
Motor
People
Table
Mean
DRBN [ 49 ]
81.8
81.1
73.5
90.2
79.6
72.6
65.1
69.9
78.2
71.2
76.7
62.7
75.2
LMT-GP(SemiLL) [ 50 ]
80.9
82.7
76.2
91.2
80.2
75.1
72.1
70.6
77.8
71.1
78.8
65.8
76.9
SemiLL + Colabator [ 10 ]
79.6
82.0
75.7
89.2
80.8
69.5
70.0
70.8
79.9
71.6
76.8
65.6
76.0
CIDNet + Colabator [ 10 ]
80.7
79.8
76.0
86.6
81.2
75.2
71.9
71.9
79.6
72.1
78.3
67.3
76.7
SemiLL + QualiTeacher
82.6
83.3
78.2
91.2
82.1
70.9
74.6
72.0
80.7
74.2
80.5
67.9
78.2
CIDNet + QualiTeacher
82.9
83.0
76.4
92.5
82.5
77.8
73.3
73.5
82.8
74.0
81.4
68.7
79.1
Table 5: Low-light image detection on ExDark .
Methods (Task: Desnowing)
CNNIQA ↑
HyperIQA ↑
ManIQA ↑
Topiq ↑
Arniqa ↑
Q-Align ↑
C2score ↑
LIQE ↑
SnowMaster(NAFNet) [ 19 ]
0.670
0.501
0.349
0.499
0.612
3.836
55.327
2.768
NAFNet + Colabator [ 10 ]
0.628
0.470
0.342
0.473
0.620
3.936
55.389
2.667
NAFNet + QualiTeacher(Topiq)
0.671
0.538
0.361
0.540
0.614
3.999
55.398
2.826
NAFNet + QualiTeacher
0.675
0.559
0.362
0.540
0.628
4.030
55.446
2.883
Table 6: Robust analysis for NR-IQA metrics. We replace MUSIQ with TOPIQ.
Real-world image restoration aims to restore high-quality (HQ) images from degraded low-quality (LQ) inputs captured under uncontrolled conditions. Existing methods typically depend on ground-truth (GT) supervision, assuming that GT provides perfect reference quality. However, GT can still contain images with inconsistent perceptual fidelity, causing models to converge to the average quality level of the training data rather than achieving the highest perceptual quality attainable. To address these problems, we propose a novel framework, termed IQPIR, that introduces an Image Quality Prior (IQP)-extracted from pre-trained No-Reference Image Quality Assessment (NR-IQA) models-to guide the restoration process toward perceptually optimal outputs explicitly. Our approach synergistically integrates IQP with a learned codebook prior through three key mechanisms: (1) a quality-conditioned Transformer, where NR-IQA-derived scores serve as conditioning signals to steer the predicted representation toward maximal perceptual quality. This design provides a plug-and-play enhancement compatible with existing restoration architectures without structural modification; and (2) a dual-branch codebook structure, which disentangles common and HQ-specific features, ensuring a comprehensive representation of both generic structural information and quality-sensitive attributes; and (3) a discrete representation-based quality optimization strategy, which mitigates over-optimization effects commonly observed in continuous latent spaces. Extensive experiments on real-world image restoration demonstrate that our method not only surpasses cutting-edge methods but also serves as a generalizable quality-guided enhancement strategy for existing methods. The code is available.
Adverse weather image restoration aims to recover images degraded by rain, haze, snow, and other weather-induced artifacts, thereby improving the robustness of outdoor vision systems. Existing unified restoration models exhibit limited generalization to real-world scenes due to their reliance on synthetic supervision and insufficient semantic constraints. In this paper, we propose a novel student--teacher semi-supervised framework that addresses both challenges. Specifically, we introduce an unreliable database that preserves failed teacher predictions as informative negative samples for contrastive learning, while a reliable database stores high-quality teacher predictions as positive samples. By jointly exploiting reliable pseudo-ground truths and unreliable teacher outputs, the proposed framework learns to enhance desirable restoration characteristics while avoiding common failures. We further propose a phase spectrum-based semantic constraint that replaces computationally expensive text-based supervision with an efficient and naturally aligned semantic prior. An adaptive phase consistency loss is also designed to dynamically balance supervision between the degraded input and teacher pseudo-ground truths according to degradation severity. Extensive experiments on real-world benchmarks demonstrate that the proposed method consistently outperforms existing state-of-the-art approaches in restoration quality and perceptual fidelity while exhibiting stronger generalization to real-world adverse weather conditions.
Cap Dang Xuan Kiet, Tat-Jen Cham
College of Computing and Data Science, Nanyang Technological University, Singapore
Image restoration aims to recover high quality images from inputs degraded by various factors, such as adverse weather, blur, or low light. While recent studies have shown remarkable progress across individual or unified restoration tasks, they still suffer from limited generalization and inefficiency when handling unknown or composite degradations. To address these limitations, we propose RAR, a Restore, Assess and Repeat process, that integrates Image Quality Assessment (IQA) and Image Restoration (IR) into a unified framework to iteratively and efficiently achieve high quality image restoration. Specifically, we introduce a restoration process that operates entirely in the latent domain to jointly perform degradation identification, image restoration, and quality verification. The resulting model is fully trainable end to end and allows for an all-in-one assess and restore approach that dynamically adapts the restoration process. Also, the tight integration of IQA and IR into a unified model minimizes the latency and information loss that typically arises from keeping the two modules disjoint, (e.g. during image and/or text decoding). Extensive experiments show that our approach consistent improvements under single, unknown and composite degradations, thereby establishing a new state-of-the-art.
I-Hsiang Chen, Isma Hadji, Enrique Sanchez +5
Samsung AI Center Cambridge · National Taiwan University · Technical University of Iasi +3