List your company
ResearchFormerly WizWand

No-Reference Image Quality Assessment on CSIQ

0.981SROCCSHAMISA
0.769880.824690.87950.93431Apr 19, 2022Dec 13, 2022Aug 9, 2023Apr 4, 2024Nov 28, 2024Jul 25, 2025Mar 21, 2026
Updated 4mo ago

Evaluation results

MethodLinks
SHAMISAType=SSL+LR2026.030.9810.987
Ours-S+Backbone=Swin-Tiny, Joint training (FR & NR tasks)=true, Architecture Type=Transformer-based2023.100.9760.977
DEFNet2025.070.9670.964
Ours-SBackbone=Swin-Tiny, Joint training (FR & NR tasks)=false, Architecture Type=Transformer-based2023.100.9640.97
QAL-IQA2025.070.9630.97
MAMIQA2026.020.9620.972
ARNIQAType=SSL+LR2026.030.9620.973
MANIQA2022.040.9610.968
MANIQAArchitecture Type=Transformer-based2023.100.9610.968
KGANet2025.070.9540.963
Ours-RBackbone=ResNet50, Joint training (FR & NR tasks)=false2023.100.9520.962
CDINet2025.070.9520.96
CausalQualityBackbone=VGG2025.070.9520.949
Re-IQA2025.070.9470.96
Re-IQAType=SSL+LR2026.030.9470.96
DBCNN2022.040.9460.959
DBCNNArchitecture Type=CNN/Hand-crafted2023.100.9460.959
DBCNN2026.020.9460.959
DB-CNNType=Supervised2026.030.9460.959
VCRNet2025.070.9430.955
CONTRIQUE2025.070.9420.955
DPNet2025.070.9420.952
CONTRIQUEType=SSL+LR2026.030.9420.955
DBCNN2025.070.940.954
CausalQualityBackbone=EffNet2025.070.9380.933
LIQE2025.070.9360.939
Su et al.Type=Supervised2026.030.9350.952
HyperIQA2025.070.9340.946
Ma192025.070.9260.929
MS-SCANet2026.020.9250.945
HyperIQA2022.040.9230.942
HyperIQAArchitecture Type=CNN/Hand-crafted2023.100.9230.942
HyperIQA2026.020.9230.942
HyperIQAType=Supervised2026.030.9230.942
TReS2022.040.9220.942
TRESArchitecture Type=CNN/Hand-crafted2023.100.9220.942
MEON2026.020.9220.942
TreS2026.020.9220.942
TReSType=Supervised2026.030.9220.942
SHAMISATraining Dataset=LIVE2026.030.909—
ARNIQATraining=LIVE2023.100.904—
ARNIQATraining Dataset=LIVE2026.030.904—
UNIQUE2025.070.9020.921
TreS2025.070.9020.923
MetaIQA2022.040.8990.908
P2P-BM2022.040.8990.902
MetaIQAArchitecture Type=CNN/Hand-crafted2023.100.8990.908
P2P-BMArchitecture Type=CNN/Hand-crafted2023.100.8990.902
TOPIQ-FR2025.070.8940.894
SHAMISATraining Dataset=KADID-10K2026.030.892—
CLIP-IQA+Model Type=Non-reasoning2026.030.890.9
CLIP-IQA+Model Category=Discriminative Deep-Learning-based, Zero-shot evaluation=true2025.090.890.9
ARNIQATraining=KADID2023.100.882—
ARNIQATraining Dataset=KADID-10K2026.030.882—
SHAMISATraining Dataset=TID20132026.030.87—
ARNIQATraining=TID20132023.100.866—
ARNIQATraining Dataset=TID20132026.030.866—
HiRQATraining Dataset=Koniq10k and HiRQA2025.080.8620.874
Re-IQATraining=KADID2023.100.855—
Re-IQATraining Dataset=KADID-10K2026.030.855—
MEON2022.040.8520.864
WaDIQaM2022.040.8520.844
MEONArchitecture Type=CNN/Hand-crafted2023.100.8520.864
WaDIQaMArchitecture Type=CNN/Hand-crafted2023.100.8520.844
Re-IQATraining=TID20132023.100.85—
Re-IQATraining Dataset=TID20132026.030.85—
DeQA-ScoreModel Type=Non-reasoning2026.030.8470.868
DeQA-ScoreModel Category=VLM-based, Zero-shot evaluation=true2025.090.8470.868
NIMAModel Type=Non-reasoning2026.030.8440.854
NIMAModel Category=Discriminative Deep-Learning-based, Zero-shot evaluation=true2025.090.8440.854
EvoQualityModel Category=VLM-based, Zero-shot evaluation=true, Evolution Round=Round 22025.090.8390.865
Su et al.Training=KADID2023.100.828—
TRIQ2026.020.8280.838
Su et al.Training Dataset=KADID-10K2026.030.828—
TIQA2022.040.8250.838
TIQAArchitecture Type=Transformer-based2023.100.8250.838
ILNIQE2022.040.8220.865
ILNIQEArchitecture Type=CNN/Hand-crafted2023.100.8220.865
BIECON2022.040.8150.823
BIECONArchitecture Type=CNN/Hand-crafted2023.100.8150.823
MUSIQModel Type=Non-reasoning2026.030.8150.86
ME-IQABase VLM=EvoQuality2026.030.8150.851
MUSIQModel Category=Discriminative Deep-Learning-based, Zero-shot evaluation=true2025.090.8150.86
BRISQUE2022.040.8120.748
BRISQUEArchitecture Type=CNN/Hand-crafted2023.100.8120.748
BRISQUE2026.020.8120.748
CONTRIQUETraining=TID20132023.100.811—
CONTRIQUETraining Dataset=TID20132026.030.811—
HyperIQATraining=KADID2023.100.809—
HyperIQATraining Dataset=KADID-10K2026.030.809—
ILNIQE2025.070.8080.851
DIIVINE2022.040.8040.776
DIIVINEArchitecture Type=CNN/Hand-crafted2023.100.8040.776
DIIVINE2026.020.8040.776
CONTRIQUETraining=LIVE2023.100.803—
CONTRIQUETraining Dataset=LIVE2026.030.803—
Re-IQATraining=LIVE2023.100.795—
Re-IQATraining Dataset=LIVE2026.030.795—
EvoQualityModel Category=VLM-based, Zero-shot evaluation=true, Evolution Round=Round 12025.090.7890.827
BaselineBase VLM=EvoQuality2026.030.7780.821