List your company
ResearchFormerly WizWand

No-Reference Image Quality Assessment on KADID-10K

0.972SROCCDistortion-Aware Fusion Framework
0.725520.789510.85350.91749Apr 19, 2022Dec 30, 2022Sep 11, 2023May 23, 2024Feb 2, 2025Oct 15, 2025Jun 28, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Distortion-Aware Fusion Framework2026.060.9720.973
B_GatedActivation=Gated2025.090.970.973
B_Gated (Ours)Activation=Gated, Type=Non-diffusion2025.090.970.973
BActivation=Baseline2025.090.9610.964
B (Baseline)Type=Non-diffusion2025.090.9610.964
LGDM2025.090.9580.961
MANIQA2026.060.9460.946
MANIQA2022.040.9440.946
Ours-S+Backbone=Swin-Tiny, Joint training (FR & NR tasks)=true, Architecture Type=Transformer-based2023.100.9440.945
DEFNet2025.070.9420.944
Ours-SBackbone=Swin-Tiny, Joint training (FR & NR tasks)=false, Architecture Type=Transformer-based2023.100.9410.942
KGANet2025.070.940.943
MANIQAArchitecture Type=Transformer-based2023.100.9390.939
B_SigActivation=Sigmoid2025.090.9390.943
B_Sig (Ours)Activation=Sigmoid, Type=Non-diffusion2025.090.9390.943
GLIANetNumber of trainable parameters=8M2026.050.9390.943
SHDIQANumber of trainable parameters=24M2026.050.9370.94
Q-Alignend-to-end fine-tuned backbone=true2026.060.9370.941
CONTRIQUE2025.070.9340.937
CONTRIQUEType=SSL+LR2026.030.9340.937
CONTRIQUE2026.060.9340.937
SLIDE-IQA-Q2026.060.9320.934
LoDAType=Non-diffusion2025.090.9310.936
LoDaNumber of trainable parameters=9M2026.050.9310.936
LIQE2025.070.930.931
DAIQA2026.050.930.925
LIQENumber of trainable parameters=151M2026.050.930.931
LIQE2026.060.930.931
SLIDE-IQA2026.060.9240.927
DPNet2025.070.9230.924
QMambaNumber of trainable parameters=50M2026.050.9230.938
SHAMISAType=SSL+LR2026.030.9220.924
CDINet2025.070.920.919
TReS2022.040.9150.858
QAL-IQA2025.070.9080.91
ARNIQAType=SSL+LR2026.030.9080.912
ARNIQA2026.060.9080.912
CausalQualityBackbone=EffNet2025.070.9070.905
CausalQualityBackbone=VGG2025.070.8990.898
TOPIQ-FR2025.070.8950.896
Ours-RBackbone=ResNet50, Joint training (FR & NR tasks)=false2023.100.8850.884
ReIQA-Q2026.060.8850.892
UNIQUE2025.070.8840.885
TreS2025.070.8810.879
DBCNN2025.070.8780.878
MUSIQType=Non-diffusion2025.090.8750.872
MUSIQ2026.050.8750.872
HyperIQA2025.070.8720.869
Re-IQA2025.070.8720.885
RE-IQAType=Non-diffusion2025.090.8720.885
Re-IQAType=SSL+LR2026.030.8720.885
Re-IQANumber of trainable parameters=48M2026.050.8720.885
MUSIQ2026.060.8720.875
ReIQA2026.060.8720.885
Su et al.Type=Supervised2026.030.8660.874
TRESArchitecture Type=CNN/Hand-crafted2023.100.8590.858
TReSType=Non-diffusion2025.090.8590.859
TReSType=Supervised2026.030.8590.858
TReS2026.050.8590.858
TReS2026.060.8590.858
VCRNet2025.070.8530.849
HyperIQA2022.040.8520.845
HyperIQAArchitecture Type=CNN/Hand-crafted2023.100.8520.845
HyperIQAType=Non-diffusion2025.090.8520.845
HyperIQAType=Supervised2026.030.8520.845
HyperIQA2026.050.8520.845
HyperIQA2026.060.8520.845
HyperIQA2026.060.8520.845
DBCNN2022.040.8510.856
DBCNNArchitecture Type=CNN/Hand-crafted2023.100.8510.856
DBCNNType=Non-diffusion2025.090.8510.856
DB-CNNType=Supervised2026.030.8510.856
DBCNN2026.050.8510.856
DBCNN2026.060.8510.856
DB-CNN2026.060.8510.856
TIQA2022.040.850.855
TIQAArchitecture Type=Transformer-based2023.100.850.855
TIQAType=Non-diffusion2025.090.850.855
P2P-BM2022.040.840.849
P2P-BMArchitecture Type=CNN/Hand-crafted2023.100.840.849
P2P-BMType=Non-diffusion2025.090.840.849
P2P-BM2026.050.840.849
CLIP-IQA+2026.060.8230.838
EvoQualityModel Category=VLM-based, Zero-shot evaluation=true, Evolution Round=Round 22025.090.8070.803
MM-IQA2026.040.8050.813
Q-AlignModel Category=VLM-based, Zero-shot evaluation=true2025.090.7920.795
EvoQualityModel Category=VLM-based, Zero-shot evaluation=true, Evolution Round=Round 12025.090.7820.784
SHAMISATraining Dataset=TID20132026.030.779—
ARNIQATraining Dataset=LIVE2026.030.764—
MetaIQA2022.040.7620.775
MetaIQAArchitecture Type=CNN/Hand-crafted2023.100.7620.775
MetaIQAType=Non-diffusion2025.090.7620.775
MetaIQA2026.050.7620.775
Q-InsightModel Category=VLM-based, Zero-shot evaluation=true2025.090.7620.759
HiRQATraining Dataset=Koniq10k and HiRQA2025.080.7610.733
CLIP-IQA+Model Category=Discriminative Deep-Learning-based, Zero-shot evaluation=true2025.090.7560.762
WaDIQaM2022.040.7390.752
WaDIQaMArchitecture Type=CNN/Hand-crafted2023.100.7390.752
WaDIQaMType=Non-diffusion2025.090.7390.752
ARNIQATraining Dataset=CSIQ2026.030.735—