List your company
ResearchFormerly WizWand

Image Quality Assessment on SPAQ

0.934SRCCDeQA-Score
0.895520.905510.91550.92549Aug 12, 2021Jun 5, 2022Mar 29, 2023Jan 20, 2024Nov 12, 2024Sep 5, 2025Jun 29, 2026
Updated 4mo ago

Evaluation results

MethodLinks
DeQA-ScoreTraining Dataset=KonIQ, SPAQ, KADID2026.010.9340.938
DeQA-ScoreTraining Dataset=KonIQ, SPAQ2026.010.9330.936
GLIANetNumber of trainable parameters=8M2026.050.9320.934
Q-Align2024.050.930.933
SHDIQANumber of trainable parameters=24M2026.050.930.932
DeQA-ScoreTraining Dataset=KonIQ, SPAQ, KADID, PIPAL2026.010.9290.932
DeQADomain Setting=In-domain2025.100.9290.932
DR.Expertspre-training=false2026.020.9280.933
LQMambapre-training=false2026.020.9270.933
MA-AGIQAType=data-driven, Supervision Setting=Supervised2026.020.9270.932
QMambaNumber of trainable parameters=50M2026.050.9270.933
QPTpre-training=true2026.020.9250.928
LODApre-training=false2026.020.9250.928
LoDa2024.110.9250.928
CSFIQA2024.110.9250.935
LoDa2024.050.9250.928
LoDaNumber of trainable parameters=9M2026.050.9250.928
QCNpre-training=false2026.020.9230.928
StairIQAType=data-driven, Supervision Setting=Supervised2026.020.9230.929
DP-IQAModel Type=teacher2024.050.9230.926
MANIQAType=data-driven, Supervision Setting=Supervised2026.020.9220.927
LIQE2024.050.9220.919
LEAFSupervision type=Weak-supervised, MOS percentage=30%2026.010.9210.922
QFM-IQMpre-training=false2026.020.920.924
QFM-IQM2024.110.920.924
Q-AlignDomain Setting=In-domain2025.100.920.917
DP-IQAModel Type=student2024.050.920.923
DEIQTpre-training=false2026.020.9190.923
DEIQT2024.110.9190.923
CDINet2024.110.9190.922
Re-IQAType=SSL + LR2023.100.9180.925
Re-IQAType=data-driven, Supervision Setting=Weak-supervised2026.020.9180.925
MUSIQ2024.110.9180.921
Re-IQA2024.110.9180.925
RACTDomain Setting=In-domain2025.100.9180.922
MUSIQ2024.050.9180.921
ReIQA2024.050.9180.925
MUSIQ2026.050.9180.921
Re-IQANumber of trainable parameters=48M2026.050.9180.925
MUSIQpre-training=false2026.020.9170.921
MUSIQ-singleMethod variant=single-scale2021.080.9170.92
MUSIQ (Ours)Method variant=multi-scale/ensemble2021.080.9170.921
HyperIQAType=Supervised learning2023.100.9160.919
HyperIQApre-training=false2026.020.9160.919
HyperIQAType=data-driven, Supervision Setting=Supervised2026.020.9150.918
ELIQType=data-driven, Supervision Setting=Weak-supervised2026.020.9150.925
DACNN2024.110.9150.921
CONTRIQUEType=SSL + LR2023.100.9140.919
CONRTIQUEpre-training=true2026.020.9140.919
CONTRIQUEType=data-driven, Supervision Setting=Weak-supervised2026.020.9140.919
CONTRIQUESupervision type=Weak-supervised2026.010.9140.919
VisualQuality-R1Domain Setting=In-domain2025.100.9140.918
DEIQT2024.050.9140.921
VisualQuality-R1Training Dataset=KADID, SPAQ2026.010.9130.917
DB-CNNType=Supervised learning2023.100.9110.915
QualiCLIP+Opinion-Unaware=false2024.030.9110.914
DBCNN2024.110.9110.915
HyperIQA2024.110.9110.915
DBCNN2021.080.9110.915
DBCNN2024.050.9110.915
HyperIQA2024.050.9110.915
Q-DeepSight†Category=MLLM (w/ reasoning), Training Dataset=KonIQ + DQ-7K2026.040.9110.912
DBCNN2026.050.9110.915
HyperIQA2026.050.9110.915
DAIQA2026.050.9110.922
CONTRIQUEOpinion-Unaware=false2024.030.910.912
DB-CNNpre-training=false2026.020.910.913
Re-IQAOpinion-Unaware=false2024.030.9090.912
DBCNNType=data-driven, Supervision Setting=Supervised2026.020.9090.927
Fang et al.Extra information usage=w/o extra info2021.080.9080.909
Q-InsightModel Category=Specialized, Domain=Cross domain2025.120.9070.905
Q-InsightDomain Setting=In-domain2025.100.9070.913
Q-InsightCategory=Specialized, Evaluation Domain=Cross-domain, Evaluation Protocol=Single-task2026.030.9070.905
Q-DeepSightCategory=MLLM (w/ reasoning), Training Dataset=KonIQ2026.040.9060.905
ARNIQAType=SSL + LR2023.100.9050.91
Q-insightTraining Dataset=KonIQ, KADIS2026.010.9050.907
Q-InsightCategory=MLLM-based, Training Dataset=KonIQ2025.100.9050.907
Q-Insight†Category=MLLM (w/ reasoning), Training Dataset=KonIQ + DQ-7K2026.040.9050.907
Tool-IQATraining Dataset=KonIQ2026.060.9050.906
ARNIQAOpinion-Unaware=false2024.030.9040.909
ARNIQAType=data-driven, Supervision Setting=Weak-supervised2026.020.9040.909
ARNIQASupervision type=Weak-supervised2026.010.9040.909
UniPerceptModel Category=Ours2025.120.9040.895
GRepQ-DType=data-driven, Supervision Setting=Weak-supervised2026.020.9030.917
Q-HawkeyeTraining Dataset=KonIQ2026.010.9030.904
Q-HawkeyeMethodology=MLLM-Based Methods2026.010.9030.904
GRepQ-DSupervision type=Weak-supervised2026.010.9030.917
Q-InsightCategory=VLM (w/o & w/ reasoning), Trained on KonIQ=true2026.010.9030.903
Q-InsightCategory=MLLM (w/ reasoning), Training Dataset=KonIQ2026.040.9030.903
Q-Insight (regression)Method Category=MLLM-based: RL training, Backbone=Qwen2.5-VL-7B, Reproduction Protocol=Reported2026.060.9030.903
GRepQOpinion-Unaware=false2024.030.9020.903
CLIP-IQA+Opinion-Unaware=false2024.030.9010.904
Zoom-IQACategory=VLM (w/o & w/ reasoning), Trained on KonIQ=true2026.010.90.902
Zoom-IQACategory=MLLM (w/ reasoning), Training Dataset=KonIQ2026.040.90.902
Tool-augmented Zoom-IQATraining Dataset=KonIQ2026.060.90.902
Q-insightVenue=NeurIPS’25, Methodology=MLLM-Based Methods2026.010.8990.903
Tool-IQATraining dataset=KADID-10K, Method category=VLM-based2026.060.8980.898
RALICategory=Non-MLLM Deep-learning, Training Dataset=KonIQ2025.100.8970.897
TATARCategory=Unified2026.030.8970.894
MR-IQAMethod Category=MLLM-based: RL training, Backbone=Qwen3-VL-2B, Reproduction Protocol=Controlled reproduction2026.060.8970.892