Awesome Perceptual Quality

Metrics, models, and subjective datasets for how images and video actually look to people.

Image quality, video quality, user-generated content, and HDR, kept in one place instead of four. Datasets sit in their own tables so you can find data by what it holds rather than by which paper introduced it. Every entry is tagged along all four axes, so an HDR video model shows up whether you came looking for HDR or for video.

Contact

Maintained by Shreshth Saini. Corrections and additions are welcome through issues or a pull request.

If the catalog is useful in your work, a star helps others find it, and you can cite it as:

@misc{saini2026awesomeperceptualquality,
  title  = {Awesome Perceptual Quality: A Tagged Catalog of Image, Video, UGC and HDR Quality Assessment},
  author = {Saini, Shreshth},
  year   = {2026},
  url    = {https://github.com/shreshthsaini/Awesome-Perceptual-Quality}
}

Tags

Tag	Meaning
	High dynamic range, wide gamut, high bit depth
	User-generated content, authentic rather than simulated distortion
	Video quality
	Still image quality

Modifiers narrow it further: FR and NR for reference availability, MLLM for language-model scorers, AIGC for generated content, SYNTHETIC and AUTHENTIC for how the distortion arose, plus GAMING, STREAMING, and COMPRESSION.

Adding a paper

Easiest way, no git at all: open an issue, fill in the form, done. A bot builds the row, validates it, checks your links, regenerates the tables and opens the pull request for you. If something is off it says so in a comment rather than failing silently.

Editing directly: README.md is generated, so change the CSVs in data/ instead, then run:

python3 scripts/validate.py
python3 scripts/generate_readme.py

Commit the CSV row and the regenerated README together. Standard library only, nothing to install.

Either way it is welcome, including your own papers. Saying you are an author just makes review faster. Conventions live in CONTRIBUTING.md.

Contents

• Datasets (63)
  • Generated content (9)  ·  diffusion and text-to-vision output
  • HDR (10)  ·  high dynamic range, wide gamut, UGC HDR
  • Video (24)  ·  UGC, streaming, gaming, compression
  • Image (20)  ·  synthetic distortion, in-the-wild, aesthetics
• Methods (115)
  • Multimodal and reasoning (24)  ·  MLLM scorers, RL-trained reasoning, benchmarks
  • HDR (15)  ·  PU encodings, VDP family, HDR UGC models
  • Video (26)  ·  full-reference, blind UGC, efficient samplers
  • No-reference image (29)  ·  NSS, CNN, transformer, CLIP
  • Full-reference image (21)  ·  classical indices, learned perceptual distances
• Challenges (8)  ·  NTIRE, AIM, AIS, CLIC, ICME, VQEG
• Surveys (7)
• Toolboxes (13)  ·  pyiqa, VMAF, PIQ, ColorVideoVDP
• Elsewhere  ·  related lists

---

Datasets

Newest first. A blank cell means the number is not reliably documented, not that it is zero.

Generated content

Subjective data for diffusion and text-to-vision output, where the failure modes differ from camera distortion.

Dataset	Content	Annotations	Venue	Tags	Links
EvalCrafter Benchmark aligning objective metrics to human opinion	700 prompts across several text-to-video models	human-aligned scores over 17 metrics	CVPR 2024	AIGC	paper
GenAI-Bench Human preference votes comparing text-to-video model outputs	video subset of GenAI-Arena outputs	1,069 pairwise human votes	NeurIPS 2024	AIGC	paper
RichHF-18K Point-level artifact and misalignment annotations, not just a single score	18,000 generated images	4 fine-grained scores plus artifact heatmaps	CVPR 2024	AIGC	paper
T2VQA-DB Large text-to-video quality set collected under an ITU-compliant protocol	10,000 videos from 9 text-to-video models	MOS from 27 annotators	ACMMM 2024	AIGC	paper
VideoFeedback Quality, consistency and alignment scores; the training set behind VideoScore	about 37,600 videos from 11 generators	multi-aspect human scores, 1 to 4	EMNLP 2024	AIGC	paper data
AGIQA-1K An early perceptual quality set for generated images, from two diffusion models	1,080 images	MOS	arXiv 2023	AIGC	paper
AGIQA-3K Spans GAN, autoregressive and diffusion text-to-image models	2,982 AI-generated images	125,244 ratings, 21 raters on 2 dimensions	TCSVT 2023	AIGC	paper
AIGCIQA2023 Six text-to-image models across 100 prompts, scored on three dimensions	2,400 images	MOS for quality, authenticity and correspondence	arXiv 2023	AIGC	paper
PKU-I2IQA Image-to-image generation quality, Midjourney and Stable Diffusion v1.5	1,600 images	32,000 ratings, 20 raters	arXiv 2023	AIGC	paper

HDR

Dataset	Content	Annotations	Venue	Tags	Links
Beyond8Bits Largest crowdsourced real-world HDR UGC quality study to date	44K videos from 6.5K sources	over 1.5M crowd ratings	CVPR 2026	AUTHENTIC	paper data
BrightVQ Subjective database dedicated to HDR UGC quality modelling	300 sources, 2,100 transcoded videos	73,794 crowd ratings	WACV 2026 Best Student Paper	AUTHENTIC	paper
CHUG Large-scale subjective study of real-world user-generated HDR quality	856 UGC HDR sources, 5,992 transcoded videos	211,848 crowdsourced ratings	ICIP 2025	AUTHENTIC	paper
HDRSDR-VQA Content-level HDR versus SDR preference measured on consumer TVs	960 videos, 54 sources, HDR and SDR pairs	over 22,000 pairwise comparisons, 145 subjects	arXiv 2025		paper data
LIVE HDR-vs-SDR HDR versus SDR preference under compression and scaling across display types	356 videos, 212 public	over 23,000 ratings, 67 subjects, 3 TV types	TIP 2024	COMPRESSION	paper data
YouTube SFV+HDR Portrait short-form HDR UGC content sourced from YouTube	300 reference videos, Rec.2020 HDR10, portrait	subjective score availability not confirmed	ICIP 2024		paper
LIVE-HDR First public HDR10 video quality database, studied under two lighting conditions	310 videos, 31 sources	about 20K ratings, 66 subjects, 2 ambient conditions	TIP 2023	COMPRESSION	paper data
SI-HDR Single-image HDR reconstruction benchmark used to stress-test HDR quality metrics	181 HDR scenes, RAW stacks plus 6 reconstruction outputs	metric benchmark rather than a full MOS study	SIGGRAPH 2022		paper data
UPIQ Merges SDR and HDR databases onto one unified quality scale	84 reference and 4,159 distorted images	MOS rescaled to an absolute photometric JOD scale	TMM 2022		paper data
MMSPG HDR JPEG-XT Classic HDR compression study used to calibrate PU and HDR-VDP metrics	20 scenes, 4 bitrates, 3 JPEG-XT profiles	MOS by DSIS, 24 subjects	QoMEX 2015	COMPRESSION	paper data

Video

Dataset	Content	Annotations	Venue	Tags	Links
GameScope Largest gaming quality set, spanning H.264, H.265 and AV1	4,048 videos from 424 clips across 74 games	about 37 ratings per video	ICIP 2026	GAMING	paper data code
KVQ Short-form UGC across capture, transcode and enhancement pipelines	4,200 videos, 600 uploaded and 3,600 processed	MOS per video	CVPR 2024		paper
TaoLive QoE Live-streaming QoE including frame-skip and variable-frame-rate distortion	1,197 videos, 42 sources	subjective QoE MOS	arXiv 2024	STREAMING	paper data
LIVE-YT-Gaming Gaming clips with mixed capture, recording and compression distortion	600 videos from 59 games, 360p to 1080p	18,600 ratings, 61 subjects	arXiv 2022	GAMING	paper data
LIVE-NFLX-II Adaptive streaming QoE driven by real network traces and client adaptation	420 videos, 15 contents	9,750 continuous and 9,750 retrospective scores	TIP 2021	STREAMING SYNTHETIC	paper data
LIVE-YT-HFR Frame rate and compression interaction, from 24 to 120 fps	480 videos, 16 contents, 1080p and 4K	19,000 ratings, 85 subjects	Access 2021	SYNTHETIC	paper data
LSVQ The largest in-the-wild UGC video quality set; underlies the PVQ models	38,811 videos, 240p to 1080p	about 5.5M patch-level scores	CVPR 2021		paper
BVI-DVC Video coding training set for CNN-based compression; 772 sequences released	800 sequences, 270p to 2160p	none; a training set, no MOS	arXiv 2020	COMPRESSION SYNTHETIC	paper data
CGVDS Cloud-gaming encodes rated for quality, fragmentation and discontinuity	270 videos, 480p to 1080p	MOS from over 100 subjects	MMSys 2020	GAMING SYNTHETIC	paper
UGC-VIDEO Short-form UGC re-encoded at several codecs and quantisation levels	50 source videos plus transcoded variants	subjective MOS	MIPR 2020	COMPRESSION	paper
Waterloo SQoE-IV Large adaptive-streaming QoE study across devices, encoders and algorithms	1,350 videos	QoE 0 to 100, 92 subjects across three device classes	arXiv 2020	STREAMING SYNTHETIC	paper data
LIVE-VQC Real in-capture UGC from 101 devices; the largest video study of its time	585 videos, 240p to 1080p	over 205,000 scores from 4,776 subjects	TIP 2019	AUTHENTIC	paper data
YouTube-UGC Creative-Commons YouTube sample built for compression and quality research	1,500 videos, 360p to 4K, 15 categories	crowdsourced MOS from over 8,000 raters	MMSP 2019	COMPRESSION	paper data
GamingVideoSET Six games at three resolutions and five bitrates, H.264 encoded	600 videos, 24 raw and 576 distorted, 1080p	subjective MOS under ITU-T P.910	NetGames 2018	GAMING SYNTHETIC	paper
LIVE-Qualcomm In-capture mobile distortion: focus, exposure and stabilisation failures	208 videos from 8 devices	MOS from 39 subjects	TCSVT 2018	AUTHENTIC	paper data
Waterloo SQoE-III Six adaptive-bitrate algorithms across 13 network conditions	470 videos, 20 references	MOS from 34 subjects	TBC 2018	STREAMING SYNTHETIC	paper data
KoNViD-1k Sampled from YFCC100M using six quality-diversity attributes	1,200 videos at 960x540	crowdsourced MOS on a 1 to 5 scale	QoMEX 2017	AUTHENTIC	paper data
LIVE-NFLX-I Handcrafted compression and rebuffering playout patterns on mobile	112 videos at 1080p	MOS from over 55 subjects	TIP 2017	STREAMING SYNTHETIC	paper data
CVD2014 Real camera in-capture distortion rather than synthetic post-processing	234 videos from 78 cameras	MOS from 210 observers	TIP 2016	AUTHENTIC	paper data
CSIQ Video Compression and transmission distortion at several severity levels	240 videos, 12 references, 832x480	DMOS from 35 subjects	JEI 2014	SYNTHETIC	paper
LIVE Mobile Mobile and tablet study of compression, packet loss, stalls and rate adaptation	210 videos, 10 references, 720p	about 5,300 scores from over 50 subjects	JSTSP 2012	SYNTHETIC	paper data
LIVE VQA The foundational video set: wireless, IP, MPEG-2 and H.264 distortion	160 videos, 10 references, 768x432	DMOS from 38 subjects	TIP 2010	SYNTHETIC	paper data
VQEG HD3 HDTV validation set encoded between 1 and 15 Mbps	72 videos, 9 sources, 1080i and 1080p	MOS from 24 subjects	VQEG 2010	SYNTHETIC	paper data
EPFL-PoliMI H.264 packet loss simulated over a noisy transmission channel	156 videos, 78 CIF and 78 4CIF	MOS from 40 subjects	QoMEX 2009	SYNTHETIC	paper data

Image

Dataset	Content	Annotations	Venue	Tags	Links
UHD-IQA Curated 4K Pixabay photos scored for aesthetic and technical quality	6,073 UHD 4K images	20 ratings per image, 10 expert raters	arXiv 2024	AUTHENTIC AESTHETIC	paper data
GFIQA-20K Large annotated face-only quality set sampled from YFCC100M	20,000 face images	MOS	TMM 2023	AUTHENTIC	paper
BIQ2021 Mixes casual, deliberately distorted and web-sourced photographs	about 12,000 images	MOS	JEI 2022	AUTHENTIC	paper
JUFE Non-uniform distortion in omnidirectional images, modelling real VR viewing	258 reference and 1,032 distorted 360 degree images	quality scores plus gaze and head-movement data	AAAI 2022	SYNTHETIC	paper
TAD66K 47 theme-specific aesthetic criteria; unusually dense per-image annotation	over 66,000 images	over 1,200 ratings per image	IJCAI 2022	AESTHETIC	paper
UID2021 Underwater image enhancement benchmark, 6 scene types by 15 algorithms	60 raw and 900 enhanced images	MOS by pairwise comparison, 52 observers	TOMM 2022	AUTHENTIC	paper
KonIQ-10k Ecologically valid in-the-wild set sampled from YFCC100M	10,073 images	1.2M ratings from 1,459 crowd workers	TIP 2020	AUTHENTIC	paper data
NNID Dedicated subjective study of night-time image quality	2,240 images across 448 scenes	MOS	TMM 2020	AUTHENTIC	paper
PaQ-2-PiQ Largest picture and patch quality study; maps patch quality onto whole pictures	40,000 images and 120,000 patches	about 4M human judgements	CVPR 2020	AUTHENTIC	paper
SPAQ Smartphone photographs with EXIF and brightness, contrast and noise labels	11,125 images	MOS plus per-attribute labels	CVPR 2020	AUTHENTIC	paper
KADID-10k 25 distortion types at 5 levels across 7 distortion categories	81 pristine and 10,125 distorted images	30 ratings per image, DMOS	QoMEX 2019	SYNTHETIC	paper data
KADIS-700k Unlabelled companion to KADID-10k for weakly supervised pretraining	140,000 pristine and 700,000 distorted images	none, deliberately unlabelled	QoMEX 2019	SYNTHETIC	paper data
Waterloo Exploration Deliberately has no MOS; tests generalisation by ranking criteria	4,744 pristine and 94,880 distorted images	none; uses D-test, L-test and P-test instead	TIP 2017	SYNTHETIC	paper data
AADB Aesthetic score alongside 11 photographer-identified attribute labels	10,000 images	5 raters per image plus 11 attributes	ECCV 2016	AESTHETIC	paper
CLIVE Mobile-captured authentic distortion, crowdsourced; the LIVE Challenge database	1,162 images	MOS, about 175 raters per image	TIP 2016	AUTHENTIC	paper data
TID2013 24 distortion types at 5 levels, from a multinational subjective study	25 reference and 3,000 distorted images	MOS from about 524k judgements, 971 observers	SPIC 2013	SYNTHETIC	paper data
AVA Photo-contest images with 66 semantic and 14 style tags	about 255,000 images	about 210 ratings per image	CVPR 2012	AESTHETIC	paper
CSIQ Five distortion types including contrast decrement and pink noise	30 reference and 866 distorted images	DMOS	JEI 2010	SYNTHETIC	paper data
TID2008 17 distortion types at 4 levels; the predecessor to TID2013	25 reference and 1,700 distorted images	MOS from about 256k judgements	AMR 2009	SYNTHETIC	paper data
LIVE IQA The foundational FR-IQA benchmark: JPEG, JPEG2000, noise, blur, fast fading	29 reference and 779 distorted images	DMOS, 20 to 25 raters per image	TIP 2006	SYNTHETIC	paper data

---

Methods

Multimodal and reasoning

Language models scoring quality, usually with an explanation attached, and the benchmarks built to test them.

Method	Paper	Venue	Tags	Code
Grounding-IQA	Grounding-IQA: Grounding Multimodal Language Model for Image Quality Assessment Ties quality description to spatial location, so the model says where the degradation is	ICLR 2026	MLLM REASONING
PreResQ-R1	PreResQ-R1: Towards Fine-Grained Rank-and-Score Reinforcement Learning for Visual Quality Assessment Disentangled preference-response GRPO unifying score regression and ranking	EMNLP 2026	MLLM RL REASONING	-
RALI	Reasoning as Representation: Rethinking Visual Reinforcement Learning in Image Quality Assessment Shows RL-trained IQA models convert visual features into aligned text representations	ICLR 2026	MLLM RL REASONING
Refine-IQA	Refine-IQA: Multi-Stage Reinforcement Finetuning for Perceptual Image Quality Assessment Multi-stage RFT with a reward-supervised think process and the Refine-Perception-20K set	AAAI 2026	MLLM RL REASONING	-
VQ-Insight	VQ-Insight: Teaching VLMs for AI-Generated Video Quality Understanding via Progressive Visual RL Progressive GRPO from image scoring to generated-video quality with multi-dimensional rewards	AAAI 2026	MLLM RL REASONING AIGC
VQAThinker	VQAThinker: Exploring Generalizable and Explainable Video Quality Assessment via Reinforcement Learning GRPO reasoning framework with bell-shaped, ranking and temporal rewards	AAAI 2026	MLLM RL REASONING
A-Bench	A-Bench: Are LMMs Masters at Evaluating AI-generated Images? Tests 18 LMMs on semantic understanding and quality of AI-generated images	ICLR 2025	MLLM BENCHMARK AIGC
DeQA-Score	Teaching Large Language Models to Regress Accurate Image Quality Scores Using Score Distribution Discretises score distributions into soft labels so MLLMs can regress accurate scores	CVPR 2025	MLLM
LMM4LMM	LMM4LMM: Benchmarking and Evaluating Large-multimodal Image Generation with LMMs All-in-one LMM metric plus the EvalMi-50K benchmark for text-to-image perception	ICCV 2025	MLLM BENCHMARK AIGC
Q-Bench-Video	Q-Bench-Video: Benchmarking the Video Quality Understanding of LMMs Spans natural, generated and CG video with open-ended and pairwise quality questions	CVPR 2025	MLLM BENCHMARK
Q-Eval	Q-Eval-100K: Evaluating Visual Quality and Alignment Level for Text-to-Vision Content 100K-instance MOS dataset and scoring model for text-to-image and text-to-video	CVPR 2025	MLLM BENCHMARK AIGC
Q-Insight	Q-Insight: Understanding Image Quality via Visual Reinforcement Learning GRPO jointly optimises score regression and degradation perception with reasoning traces	NeurIPS 2025	MLLM RL REASONING
Q-Ponder	Q-Ponder: A Unified Training Pipeline for Reasoning-based Visual Quality Assessment Cold-start SFT then GRPO, jointly optimising quality scoring and reasoning consistency	arXiv 2025	MLLM RL REASONING
VisualQuality-R1	VisualQuality-R1: Reasoning-Induced Image Quality Assessment via Reinforcement Learning to Rank GRPO with Thurstone comparative rewards; an early open reasoning NR-IQA model	NeurIPS 2025	MLLM RL REASONING
AesExpert	AesExpert: Towards Multi-modality Foundation Model for Image Aesthetics Perception AesMMIT instruction set of 409K records for aesthetic perception	ACMMM 2024	MLLM AESTHETIC
Co-Instruct	Towards Open-Ended Visual Quality Comparison Multi-image quality comparison with MICBench; beats GPT-4V on open-ended comparison	ECCV 2024	MLLM
Compare2Score	Adaptive Image Quality Assessment via Teaching Large Multimodal Model to Compare Turns pairwise MOS comparisons into continuous scores by soft comparison and MAP	NeurIPS 2024	MLLM
DepictQA	Depicting Beyond Scores: Advancing Image Quality Assessment through Multi-modal Language Models Language-based comparative and descriptive IQA rather than score-only assessment	ECCV 2024	MLLM REASONING
Q-Align	Q-Align: Teaching LMMs for Visual Scoring via Discrete Text-Defined Levels Unifies image quality, aesthetics and video scoring through text-defined rating levels	ICML 2024	MLLM
Q-Bench	Q-Bench: A Benchmark for General-Purpose Foundation Models on Low-level Vision Evaluates MLLMs on low-level perception, description and quality assessment	ICLR 2024	MLLM BENCHMARK
Q-Boost	Q-Boost: On Visual Quality Assessment Ability of Low-level Multi-Modality Foundation Models Triadic-tone prompting and multi-prompt ensembling to lift MLLM scoring accuracy	ICMEW 2024	MLLM	-
Q-Ground	Q-Ground: Image Quality Grounding with Large Multi-modality Models Fine-grained quality grounding over image, text and distortion-mask triplets	ACMMM 2024	MLLM REASONING
Q-Instruct	Q-Instruct: Improving Low-level Visual Abilities for Multi-modality Foundation Models 200K instruction-tuning set built from 58K human low-level visual feedback records	CVPR 2024	MLLM
VideoScore	VideoScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generation VideoFeedback set of 37.6K videos; reports 77.1 Spearman correlation with humans	EMNLP 2024	MLLM BENCHMARK AIGC

HDR

Luminance range and bit depth change what the distortions look like, so SDR metrics transfer badly.

Method	Paper	Venue	Tags	Code
BrightRate	BrightRate: Quality Assessment for User-Generated HDR Videos CONTRIQUE, CLIP and HDR luminance features regressed to MOS for UGC HDR video	WACV 2026 Best Student Paper	NR DEEP
HDR-Q	Seeing Beyond 8bits: Subjective and Objective Quality Assessment of HDR-UGC Videos First multimodal-LLM no-reference quality model for HDR user-generated video	CVPR 2026	NR MLLM DEEP
Blind HDR-IQA	Blind HDR Image Quality Assessment Based on Aggregating Perception and Inference Features Multi-scale Retinex perceptual features fused with VGG16 deep features	Sci Rep 2025	NR DEEP	-
CompressedVQA-HDR	CompressedVQA-HDR: Generalized Full-Reference and No-Reference Quality Assessment Models for Compressed HDR Videos Swin-Transformer FR and SigLip2 NR models; won the ICME 2025 HDR/SDR challenge FR track	ICMEW 2025	FR NR DEEP COMPRESSION
Cut-FUNQUE	Cut-FUNQUE: An Objective Quality Model for Compressed Tone-Mapped HDR Videos Extends FUNQUE to compressed, tone-mapped HDR video pipelines	SPIC 2025	FR CLASSICAL COMPRESSION	-
ColorVideoVDP	ColorVideoVDP: A Visual Difference Predictor for Image, Video and Display Distortions Chromatic and achromatic spatiotemporal metric covering HDR and AR/VR display distortions	TOG 2024	FR CLASSICAL
FUNQUE+	A FUNQUE Approach to the Quality Assessment of Compressed HDR Videos Wavelet-domain shared-feature framework, more accurate and cheaper than VMAF on HDR	PCS 2024	FR CLASSICAL COMPRESSION	-
HDR-ChipQA	HDR-ChipQA: No-Reference Quality Assessment on High Dynamic Range Videos NSS space-time chips with an expansive nonlinearity that emphasises luminance extremes	SPIC 2024	NR CLASSICAL	-
HIDRO-VQA	HIDRO-VQA: High Dynamic Range Oracle for Video Quality Assessment Self-supervised contrastive fine-tuning transfers SDR quality features into the HDR domain	WACVW 2024	NR DEEP
HDR-VDP-3	HDR-VDP-3: A Multi-Metric for Predicting Image Differences, Quality and Contrast Distortions in High Dynamic Range and Regular Content Unified visibility, quality and contrast-distortion metric spanning HDR and SDR	arXiv 2023	FR CLASSICAL	link
FovVideoVDP	FovVideoVDP: A Visible Difference Predictor for Wide Field-of-View Video Foveated spatio-temporal CSF metric for wide field-of-view, HDR and VR displays	TOG 2021	FR CLASSICAL
PU21	PU21: A Novel Perceptually Uniform Encoding for Adapting Existing Quality Metrics for HDR Revised PU encoding fit to 10,000 cd/m2, underlying PU-PSNR, PU-SSIM and PU-FSIM	PCS 2021	FR PU CLASSICAL
HDR-VQM	HDR-VQM: An Objective Quality Measure for High Dynamic Range Video Spatio-temporal frequency decomposition; one of the first HDR video quality measures	SPIC 2015	FR CLASSICAL
HDR-VDP-2	HDR-VDP-2: A Calibrated Visual Metric for Visibility and Quality Predictions in All Luminance Conditions Visual-system model calibrated across the luminance range, predicting visibility and quality	TOG 2011	FR CLASSICAL	link
PU encoding	Extending Quality Metrics to Full Luminance Range Images The original perceptually uniform encoding that lets PSNR and SSIM run on HDR luminance	HVEI 2008	FR PU CLASSICAL	-

Video

Temporal quality, including the UGC case where authentic distortions stack on top of each other.

Method	Paper	Venue	Tags	Code
FineVQ	FineVQ: Fine-Grained User Generated Content Video Quality Assessment Fine-grained UGC scoring across multiple dimensions with quality-aware LMM reasoning	CVPR 2025	NR DEEP MLLM
KSVQE	KVQ: Kwai Video Quality Assessment for Short-form Videos Short-form UGC scoring with CLIP-based content and distortion understanding	CVPR 2024	NR DEEP	-
MinimalisticVQA	Analysis of Video Quality Datasets via Design of Minimalistic Video Quality Models Probes dataset bias using deliberately bare-bones models built from minimal parts	TPAMI 2024	NR DEEP
ModularBVQA	Modular Blind Video Quality Assessment Modular design that explicitly models resolution and frame-rate sensitivity	CVPR 2024	NR DEEP
T2VQA	Subjective-Aligned Dataset and Metric for Text-to-Video Quality Assessment Text-to-video quality metric trained on the 10K-video T2VQA-DB	ACMMM 2024	NR DEEP AIGC
Ada-DQA	Ada-DQA: Adaptive Diverse Quality-aware Feature Acquisition for Video Quality Assessment Adaptively pools diverse quality-aware features from frozen pretrained backbones	ACMMM 2023	NR DEEP	-
DOVER	Exploring Video Quality Assessment on User Generated Contents from Aesthetic and Technical Perspectives Separates aesthetic from technical quality, with the DIVIDE-3k database	ICCV 2023	NR DEEP AESTHETIC
FasterVQA	Neighbourhood Representative Sampling for Efficient End-to-end Video Quality Assessment Extends FAST-VQA with spatial-temporal grid sampling for a large efficiency gain	TPAMI 2023	NR DEEP
Zoom-VQA	Zoom-VQA: Patches, Frames and Clips Integration for Video Quality Assessment Integrates patch, frame and clip levels; second place in the NTIRE 2023 challenge	CVPRW 2023	NR DEEP
BVQA-PT	Blindly Assess Quality of In-the-Wild Videos via Quality-aware Pre-training and Motion Perception Transfers IQA pretraining and action-recognition motion features into blind VQA	TCSVT 2022	NR DEEP
FAST-VQA	FAST-VQA: Efficient End-to-end Video Quality Assessment with Fragment Sampling Fragment-sampling transformer that cuts compute by orders of magnitude	ECCV 2022	NR DEEP TRANSFORMER
SimpleVQA	A Deep Learning based No-reference Quality Assessment Model for UGC Videos End-to-end spatial features plus low-resolution motion, deliberately minimal	ACMMM 2022	NR DEEP
MDTVSFA	Unified Quality Assessment of In-the-Wild Videos with Mixed Datasets Training Mixed-dataset training with perceptual scale alignment across four UGC databases	IJCV 2021	NR DEEP
PVQ	Patch-VQ: Patching Up the Video Quality Problem Local-to-global space-time patch model trained on the LSVQ database	CVPR 2021	NR DEEP
RAPIQUE	RAPIQUE: Rapid and Accurate Video Quality Prediction of User Generated Content Pairs quality-aware scene statistics with CNN semantics for fast UGC scoring	OJSP 2021	NR CLASSICAL DEEP
ST-GREED	ST-GREED: Space-Time Generalized Entropic Differences for Frame Rate Dependent Video Quality Prediction Entropic model capturing how frame rate and compression interact in high-frame-rate video	TIP 2021	FR CLASSICAL
VIDEVAL	UGC-VQA: Benchmarking Blind Video Quality Assessment for User Generated Content Feature-selected fusion of 60 handcrafted features, plus a UGC benchmarking study	TIP 2021	NR CLASSICAL
CNN-TLVQM	Blind Natural Video Quality Prediction via Statistical Temporal Features and Deep Spatial Features Hybrid of handcrafted temporal statistics and deep spatial CNN features	ACMMM 2020	NR CLASSICAL DEEP
ST-VMAF	Spatiotemporal Feature Integration and Model Fusion for Full Reference Video Quality Assessment Spatiotemporal and ensemble VMAF extensions adding temporal features to the fusion	TCSVT 2019	FR DEEP
TLVQM	Two-Level Approach for No-Reference Consumer Video Quality Assessment Two-layer low and high complexity features aimed at consumer capture artefacts	TIP 2019	NR CLASSICAL
VSFA	Quality Assessment of In-the-Wild Videos Content-aware CNN features with GRU temporal pooling and a hysteresis model	ACMMM 2019	NR DEEP
SpEED-QA	SpEED-QA: Spatial Efficient Entropic Differencing for Image and Video Quality Spatial-domain entropic differencing, far cheaper than ST-RRED	SPL 2017	FR RR CLASSICAL
VIIDEO	A Completely Blind Video Integrity Oracle Opinion- and distortion-unaware, using only intrinsic statistical regularities	TIP 2016	NR CLASSICAL
VMAF	Toward A Practical Perceptual Video Quality Metric Fusion-based full-reference metric; the de facto industry standard for streaming	Netflix 2016	FR DEEP COMPRESSION
V-BLIINDS	Blind Prediction of Natural Video Quality DCT-domain spatiotemporal statistics with a motion coherence model	TIP 2014	NR CLASSICAL
ST-RRED	Video Quality Assessment by Reduced Reference Spatio-Temporal Entropic Differencing Entropic differencing of spatial and temporal wavelet coefficients	TCSVT 2013	FR RR CLASSICAL

No-reference image

Blind scoring, which is what most deployments need since the reference is rarely there.

Method	Paper	Venue	Tags	Code
FGResQ	Fine-grained Image Quality Assessment for Perceptual Image Restoration Fine-grained NR-IQA model with the FGRestore dataset, targeting restored-image quality	AAAI 2026	NR DEEP
ARNIQA	ARNIQA: Learning Distortion Manifold for Image Quality Assessment Self-supervised contrastive encoder learns a distortion manifold for blind scoring	WACV 2024	NR DEEP
BIQA-CL	Task-Specific Normalization for Continual Learning of Blind Image Quality Models Continual learning across IQA datasets via task-specific batch normalisation	TIP 2024	NR DEEP
GRepQ	Learning Generalizable Perceptual Representations for Data-Efficient No-Reference IQA Self-supervised CLIP encoders contrast quality groups for few-shot NR-IQA	WACV 2024	NR CLIP
QCN	Blind Image Quality Assessment Based on Geometric Order Learning Orders feature vectors by quality using comparison transformers and learned score pivots	CVPR 2024	NR DEEP TRANSFORMER
QualiCLIP	Quality-Aware Image-Text Alignment for Opinion-Unaware Image Quality Assessment Quality-aware CLIP fine-tuning with antonym prompts, needs no MOS labels	arXiv 2024	NR CLIP
TOPIQ	TOPIQ: A Top-Down Approach From Semantics to Distortions for Image Quality Assessment Semantic attention guides distortion localisation; ships both FR and NR variants	TIP 2024	NR FR DEEP TRANSFORMER
CLIP-IQA	Exploring CLIP for Assessing the Look and Feel of Images Zero-shot quality and abstract-perception scoring via CLIP antonym prompt pairs	AAAI 2023	NR CLIP
DEIQT	Data-Efficient Image Quality Assessment with Attention-Panel Decoder Transformer attention-panel decoder reaches strong accuracy from limited labels	AAAI 2023	NR DEEP TRANSFORMER
LIQE	Blind Image Quality Assessment via Vision-Language Correspondence: A Multitask Learning Perspective Multitask CLIP correspondence across scene, distortion and quality-level text prompts	CVPR 2023	NR CLIP TRANSFORMER
Re-IQA	Re-IQA: Unsupervised Learning for Image Quality Assessment in the Wild Dual contrastive encoders separate content from quality-aware unsupervised features	CVPR 2023	NR DEEP
CONTRIQUE	Image Quality Assessment Using Contrastive Learning Contrastive pretraining predicts distortion type and degree, then linear-maps to quality	TIP 2022	NR DEEP
MANIQA	MANIQA: Multi-dimension Attention Network for No-Reference Image Quality Assessment ViT features with channel and spatial attention; won the NTIRE 2022 NR track	CVPRW 2022	NR DEEP TRANSFORMER
TReS	No-Reference Image Quality Assessment via Transformers, Relative Ranking, and Self-Consistency CNN and transformer branches with relative-ranking and self-consistency losses	WACV 2022	NR DEEP TRANSFORMER
MUSIQ	MUSIQ: Multi-scale Image Quality Transformer Multi-scale transformer handles native resolution and aspect ratio without cropping	ICCV 2021	NR DEEP TRANSFORMER
UNIQUE	Uncertainty-Aware Blind Image Quality Assessment in the Laboratory and Wild Unified model trained across several IQA datasets with pairwise learning and uncertainty	TIP 2021	NR DEEP
DBCNN	Blind Image Quality Assessment Using a Deep Bilinear Convolutional Neural Network Bilinear pooling fuses synthetic-distortion and authentic-distortion CNN features	TCSVT 2020	NR DEEP
HyperIQA	Blindly Assess Image Quality in the Wild Guided by a Self-Adaptive Hyper Network Hypernetwork generates content-adaptive perception rules for authentic distortion	CVPR 2020	NR DEEP
MetaIQA	MetaIQA: Deep Meta-Learning for No-Reference Image Quality Assessment Meta-learns a shared distortion prior, then fast-adapts to unseen distortion types	CVPR 2020	NR DEEP
PaQ-2-PiQ	From Patches to Pictures (PaQ-2-PiQ): Mapping the Perceptual Space of Picture Quality Region-based model trained on 40K real images with 4M patch quality judgements	CVPR 2020	NR DEEP AUTHENTIC
FRIQUEE	Perceptual Quality Prediction on Authentically Distorted Images Using a Bag of Features Approach A large bag of NSS features across colour spaces, aimed at authentic camera distortion	JoV 2017	NR CLASSICAL AUTHENTIC
RankIQA	RankIQA: Learning from Rankings for No-Reference Image Quality Assessment Pretrains a Siamese ranking network on synthetically ordered distortion pairs	ICCV 2017	NR DEEP
HOSA	Blind Image Quality Assessment Based on High Order Statistics Aggregation Small codebook with soft-assigned mean, variance and skewness aggregation	TIP 2016	NR CLASSICAL	-
IL-NIQE	A Feature-Enriched Completely Blind Image Quality Evaluator Opinion-unaware model adding gradient, colour and log-Gabor statistics to NIQE	TIP 2015	NR CLASSICAL	link
NIQE	Making a Completely Blind Image Quality Analyzer Opinion- and distortion-unaware: needs no human ratings and no distorted training images	SPL 2013	NR CLASSICAL
BLIINDS-II	Blind Image Quality Assessment: A Natural Scene Statistics Approach in the DCT Domain NSS model of local DCT coefficients with a simple probabilistic quality predictor	TIP 2012	NR CLASSICAL
BRISQUE	No-Reference Image Quality Assessment in the Spatial Domain Spatial-domain NSS features from locally normalised luminance, no transform required	TIP 2012	NR CLASSICAL
CORNIA	Unsupervised Feature Learning Framework for No-Reference Image Quality Assessment Codebook learned from raw patches by k-means, soft-assignment encoding, max pooling	CVPR 2012	NR CLASSICAL
DIIVINE	Blind Image Quality Assessment: From Natural Scene Statistics to Perceptual Quality Two stages: identify the distortion type, then apply distortion-specific scoring	TIP 2011	NR CLASSICAL

Full-reference image

Method	Paper	Venue	Tags	Code
JLD	JLD: Perceptual Distance Through A Jacobian Lens Frozen encoder Jacobian defines a perceptual distance fitted once from unlabeled images	arXiv 2026	FR DEEP TRANSFORMER
A-FINE	Toward Generalized Image Quality Assessment: Relaxing the Perfect Reference Quality Assumption Adaptively blends fidelity and naturalness so the reference need not be flawless	CVPR 2025	FR DEEP
MILO	MILO: A Lightweight Perceptual Quality Metric for Image and Latent-Space Optimization Compact multiscale FR metric trained on pseudo-MOS labels, fast enough to use as a loss	arXiv 2025	FR DEEP	-
ZS-IQA	Foundation Models Boost Low-Level Perceptual Similarity Metrics Zero-shot distances between intermediate DINO and CLIP ViT features beat trained deep metrics	ICASSP 2025	FR DEEP
DreamSim	DreamSim: Learning New Dimensions of Human Visual Similarity using Synthetic Data Mid-level metric tuned on synthetic triplets, sensitive to layout and semantics, not just pixels	NeurIPS 2023	FR DEEP
AHIQ	Attentions Help CNNs See Better: Attention-based Hybrid Image Quality Assessment Network ViT and CNN hybrid with deformable convolution; won the NTIRE 2022 FR-IQA challenge	CVPRW 2022	FR DEEP TRANSFORMER
DISTS	Image Quality Assessment: Unifying Structure and Texture Similarity Combines structure and texture correlations, tolerant to texture resampling	TPAMI 2020	FR DEEP
LPIPS	The Unreasonable Effectiveness of Deep Features as a Perceptual Metric Calibrated deep feature distance; still the standard learned perceptual baseline	CVPR 2018	FR DEEP
PieAPP	PieAPP: Perceptual Image-Error Assessment through Pairwise Preference Trains on pairwise preference probabilities instead of absolute quality scores	CVPR 2018	FR DEEP
WaDIQaM	Deep Neural Networks for No-Reference and Full-Reference Image Quality Assessment Patch-wise CNN jointly learns local quality scores and local pooling weights	TIP 2018	FR NR DEEP
DeepQA	Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework CNN learns a visual-sensitivity weighting map directly from IQA database statistics	CVPR 2017	FR DEEP	-
NLPD	Perceptual Image Quality Assessment Using a Normalized Laplacian Pyramid Laplacian pyramid with local gain control mimics early visual-system nonlinearities	HVEI 2016	FR CLASSICAL
GMSD	Gradient Magnitude Similarity Deviation: A Highly Efficient Perceptual Image Quality Index Pixel-wise gradient similarity with standard-deviation pooling; fast and competitive	TIP 2014	FR CLASSICAL	link
VSI	VSI: A Visual Saliency-Induced Index for Perceptual Image Quality Assessment Uses visual saliency both as a local quality feature and as a pooling weight	TIP 2014	FR CLASSICAL	-
FSIM	FSIM: A Feature Similarity Index for Image Quality Assessment Phase congruency as the primary feature, gradient magnitude as the secondary feature	TIP 2011	FR CLASSICAL	link
MAD	Most Apparent Distortion: Full-Reference Image Quality Assessment and the Role of Strategy Combines detection-based and appearance-based strategies across distortion severities	JEI 2010	FR CLASSICAL	-
PSNR	Scope of Validity of PSNR in Image/Video Quality Assessment The MSE-based baseline; this paper maps where PSNR does and does not track perception	EL 2008	FR CLASSICAL	-
VSNR	VSNR: A Wavelet-Based Visual Signal-to-Noise Ratio for Natural Images Wavelet contrast-threshold model of near- and supra-threshold distortion visibility	TIP 2007	FR CLASSICAL	-
VIF	Image Information and Visual Quality Models source and distortion as information channels; scores shared image information	TIP 2006	FR CLASSICAL	link
SSIM	Image Quality Assessment: From Error Visibility to Structural Similarity Structural similarity across local luminance, contrast and structure; the field's reference point	TIP 2004	FR CLASSICAL	link
MS-SSIM	Multiscale Structural Similarity for Image Quality Assessment Multiscale extension of SSIM, calibrated across scales for varying viewing conditions	Asilomar 2003	FR CLASSICAL	link

---

Challenges

Challenge	Years	What it covers
NTIRE Perceptual IQA Challenge	2021-2022	CVPR NTIRE challenge on full-reference and no-reference IQA for GAN-restored images
NTIRE Quality Assessment of AI-Generated Content	2024-2025	CVPR NTIRE challenge benchmarking IQA and VQA on AI-generated images and video
NTIRE Short-form UGC Video Quality Assessment	2024-2025	CVPR NTIRE challenge on efficient VQA for short-form user-generated video
AIM Compressed Video Quality Assessment	2024	ECCV AIM challenge evaluating VQA models across 14 codecs and standards
AIS Video Quality Assessment of UGC	2024	CVPR AI for Streaming challenge on blind VQA for user-generated video
CLIC: Challenge on Learned Image Compression	2018-2025	Annual workshop challenge benchmarking learned image and video compression quality
ICME Generalizable HDR and SDR Video Quality Measurement	2025	ICME grand challenge benchmarking VQA models across both HDR and SDR content
VQEG	1997-ongoing	Cross-industry body running ongoing subjective and objective video quality projects

Surveys

Survey	Venue	Summary
A Survey on Image Quality Assessment: Insights, Analysis, and Future Outlook	arXiv 2025	Statistical, ML and deep IQA methods with future directions
Quality Assessment in the Era of Large Models: A Survey	arXiv 2024	Quality assessment performed by, and applied to, large multimodal models
Video Quality Assessment: A Comprehensive Survey	arXiv 2024	Broad survey of VQA algorithms, benchmarks and databases including LMM-based methods
Perceptual Video Quality Assessment: A Survey	SCIS 2024	Review of subjective VQA methodology alongside objective VQA models
A Survey of AI-Generated Video Evaluation	arXiv 2024	Metric-based, human-involved and model-centered evaluation of generated video
Blind Image Quality Assessment: A Brief Survey	arXiv 2023	Compact survey of hand-crafted, deep and multimodal no-reference IQA
Perceptual Image Quality Assessment: A Survey	SCIS 2020	Foundational review of IQA databases and classical to early-deep methods

Toolboxes

Working implementations. Worth preferring over a fresh reimplementation, since small preprocessing differences move scores more than people expect.

Toolbox	Stars	Language	What it does
FFmpeg		C	Includes libvmaf, SSIM and PSNR quality filters for video encoding pipelines
Netflix VMAF (libvmaf)		C	Perceptual video quality metric based on multi-method fusion, includes libvmaf
LPIPS		Python	Reference implementation of Learned Perceptual Image Patch Similarity, pip install lpips
IQA-PyTorch (pyiqa)		Python	PyTorch toolbox for IQA: PSNR, SSIM, LPIPS, NIQE, MUSIQ, TOPIQ, NIMA, BRISQUE and more
TorchMetrics		Python	PyTorch metrics library with an image module covering SSIM, LPIPS, FID, PSNR
PIQ		Python	Collection of measures and metrics for image-to-image tasks in PyTorch
sewar		Python	Python package bundling PSNR, SSIM, MS-SSIM, VIFp, UQI and other full-reference IQA metrics
DISTS		Python	Reference implementation of the Deep Image Structure and Texture Similarity metric
SPORCO		Python	Sparse optimization research code used in restoration and quality-related pipelines
ColorVideoVDP		Python	Perceptual difference metric for color, luminance and HDR images/video
HDRutils		Python	Utilities for HDR image merging, I/O and processing used in HDR quality pipelines
MSU VQMT SDK		C++	SDK for MSU Video Quality Measurement Tool; build custom video/image quality plugins
JLD		Python	Full-reference image metric with a fitted Jacobian lens, local distance maps and differentiable loss

Elsewhere

• Awesome-Image-Quality-Assessment, the reference IQA list, paired with the IQA-PyTorch toolbox
• Awesome-High-Dynamic-Range-Imaging, HDR reconstruction and imaging
• Awesome-Aesthetic-Evaluation-and-Cropping, aesthetics rather than fidelity

License

CC0. The catalog metadata is public domain. Linked papers, datasets, and code stay under their own licenses.