Pith. sign in

REVIEW 3 major objections 6 minor 80 references

Quality Assessment and Distortion-aware Saliency Prediction for AI-Generated Omnidirectional Images

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that AI-generated 360° images can be automatically evaluated from three human-experience perspectives and repaired by using predicted distortion-aware saliency to guide inpainting.

desk verdict Useful new database for AIGODI quality, but the distortion-aware saliency ground truth is under-specified and unvalidated, which is the main thing to fix before trusting the saliency and optimization claims. read the letter →

arxiv 2506.21925 v1 pith:YQ4N6ENO submitted 2025-06-27 cs.CV

Liu Yang , Huiyu Duan , Jiarui Wang , Jing Liu , Menghan Hu , Xiongkuo Min , Guangtao Zhai , Patrick Le Callet This is my paper
classification cs.CV
keywords AI-generatedomnidirectionalimagesimagequalityassessmentdistortion-awaresaliencypredictionhumanvisualexperienceBLIP-2generationsaliency-guidedinpaintingOHF2024database
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that the quality of AI-generated omnidirectional (360°) images can be assessed and improved by modeling human feedback directly. To support that, the authors built OHF2024, a database of 600 generated 360° images with subjective scores on three dimensions—quality, comfortability, and text-image correspondence—plus 600 distortion-aware saliency maps marking the regions humans find most degraded. On top of this database, they trained two models that share a BLIP-2 encoder: BLIP2OIQA, which predicts the three scores, and BLIP2OISal, which predicts distortion-aware saliency. They report state-of-the-art results against 21 image-quality models and 20 saliency models, and they show that using the predicted saliency to guide inpainting raises both machine-predicted and human-rated quality. If the claims hold, automatic pipelines could filter, repair, and curate AI-generated immersive content for VR and AR.

What carries the argument

The load-bearing object is the shared-wise encoder built on BLIP-2: a frozen Q-Former interacting with partially frozen image and text encoders produces text-image fused features, while five layers of image features are kept for spatial detail. On top of it, BLIP2OIQA uses a viewport-split step (six 110° viewports), then self-attention for intra-viewport quality and three cross-attention aggregators to form perspective-aware representations for quality, comfortability, and correspondence, regressed by MLPs. BLIP2OISal uses a feature-fusion module that combines the last two image-feature layers with the text-image fused feature through a spatial transformer, followed by three hierarchical feature-refine modules that progressively merge earlier image features at multiple scales. The quality model is trained with L1 loss against MOS; the saliency model is trained with a weighted combination of a correlation loss and a KL-divergence loss.

What would settle it

Collect the raw click annotations, measure inter-subject agreement (for example, overlap of click centers), and retrain BLIP2OISal on maps built with a different aggregation rule such as majority voting instead of smoothed centers; if performance on held-out human maps collapses or changes substantially across aggregation rules, the saliency ground truth and the model trained on it are not stable.

Watch

Extended reading notes

Core claim

The central discovery is that a single pretrained vision-language encoder (BLIP-2) can support both tasks needed to make AI-generated 360° images usable: predicting how humans will experience them, and locating the distorted regions that spoil that experience. BLIP2OIQA splits an omnidirectional image into six viewports, fuses each with the text prompt, and applies self-attention within viewports and cross-attention across viewports to produce separate scores for quality, comfortability, and correspondence; the paper reports SRCC values of 0.9074, 0.8763, and 0.8354 on these dimensions. BLIP2OISal fuses text-image features with multi-layer spatial image features in a saliency decoder and reports AUC 0.8312 and CC 0.4324 on distortion-aware saliency. The same predicted saliency, thresholded into viewport masks, guides an inpainting-based optimization that improves human ratings (for example, quality rises from 46.31 to 50.66 on low-quality images) and moves the whole loop toward automatic quality control for AI-generated omnidirectional content.

Load-bearing premise

The load-bearing premise is that the distortion-aware saliency maps derived from subjects' handle-click directions reliably mark the regions that are both salient and distorted; no inter-subject agreement or external validation of this click-to-map protocol is reported.

Editorial extensions

If this is right

  • The OHF2024 database gives the community 600 images, 1,800 MOS ratings across three dimensions, and 600 distortion-aware saliency maps, all to be released.
  • BLIP2OIQA can rank AI-generated 360° images on quality, comfortability, and prompt correspondence better than the 21 compared no-reference IQA models, including on a train/test split that separates generation models.
  • BLIP2OISal can mark distortion-aware salient regions better than the 20 compared saliency models on the new task.
  • Saliency-guided inpainting of predicted distorted viewports improves both predicted and human-rated scores on all three dimensions for low- and high-quality images.
  • The two models share a single encoder, so one forward pass can produce both an experience profile and an edit mask for an image.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the click-based annotation protocol were cross-checked with eye-tracking, the distortion-aware saliency task could be benchmarked against established fixation datasets and the click-to-map aggregation could be tested for subject bias.
  • Because distortion-aware saliency is tied to the text prompt, the same shared-encoder architecture could be adapted for prompt-conditional quality control during generation, not just post-hoc inpainting.
  • The three rating dimensions are only weakly correlated, especially correspondence versus the other two, so general-purpose IQA models that collapse quality to one score may miss what makes AI-generated omnidirectional images fail for immersive use.
  • A testable extension is to apply the optimization loop to other 360° generators and measure whether inpainting gains persist when the repair model differs from the generation model.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper introduces OHF2024, a new database of 600 AI-generated omnidirectional images (AIGODIs) with subjective ratings from three dimensions (quality, comfortability, and correspondence) and distortion-aware saliency maps derived from human clicks. It proposes two models sharing a BLIP-2 based encoder: BLIP2OIQA for multi-perspective quality assessment and BLIP2OISal for distortion-aware saliency prediction, and an automatic optimization pipeline that uses predicted scores and saliency maps to guide viewport-based inpainting. Experiments report state-of-the-art performance on both tasks (e.g., BLIP2OIQA SRCC 0.9074 for quality in Table I; BLIP2OISal AUC 0.8312 in Table III) and improved human ratings after optimization (Table VII). The database and code are promised for release.

Significance. If the claims hold, the paper fills a clear gap: no existing database or models jointly address quality assessment, distortion-aware saliency, and optimization for AIGODIs. The three-dimensional evaluation (quality, comfortability, correspondence) is a thoughtful adaptation of general AIGC-IQA to the immersive, text-prompt-driven ODI setting. The proposed shared-encoder architecture with task-specific decoders is reasonable and the optimization demonstration is a useful practical application. The planned release of the database and code is a concrete community contribution that will enable further work. However, the significance is conditional on the reliability of the novel distortion-aware saliency ground truth and on the statistical robustness of the reported performance gains.

major comments (3)
  1. [Section III-C2] The construction of the distortion-aware saliency ground truth is under-specified and unvalidated. The text says subjects provide 'handle-click direction' for regions with severe distortions, and then 'we first label the center of the distorted salient regions in an image, then the fixation maps are smoothed with a 0.4° Gaussian kernel.' It does not report how clicks from 20 subjects were aggregated into one map, whether multiple clicks per image were allowed, how a 'distorted salient region' was distinguished from any distorted region, or any inter-subject agreement statistic. No external validation (e.g., eye-tracking, second annotation pass, or expert localization) is provided. Since Table III and the optimization results in Section VI rely directly on these maps, the saliency and optimization claims inherit this risk. Please provide the full annotation protocol, aggregation method, agreement measures, and a validation experiment.
  2. [Section V-A and Tables I-III] All reported performance numbers for BLIP2OIQA, BLIP2OISal, and the baselines are single point estimates from one random 7:3 scene-based split. No standard deviations, confidence intervals, or significance tests are given for the SRCC/PLCC/KRCC comparisons. The ROC analysis in Fig. 11 is not fully described: the construction of the image pairs, the number of trials, and the statistical test behind the significance matrices are missing. Without multiple splits or bootstrapping, the claimed SOTA improvements (e.g., Table I, quality SRCC 0.9074 vs. 0.8793) could be within noise. Please add multiple splits or significance testing for both the IQA and saliency results.
  3. [Section VI-B and Table VII] The optimization process introduces a free threshold x0 in Eq. (14) to convert saliency maps into inpainting masks, but no sensitivity analysis is provided; the reported improvements in Table VII may depend strongly on its value. The subjective study for Table VII is not described in terms of number of subjects, whether they were different from the database collection subjects, or per-image variance. In addition, because the low-quality/high-quality grouping and the post-optimization evaluation both use BLIP2OIQA, there is a mild circularity; the human ratings help address this, but without error bars it is unclear whether the differences between 'Before' and 'After' (e.g., Quality 46.31 to 50.66) are statistically significant. Please add threshold sensitivity, subject details, and statistical testing.
minor comments (6)
  1. [Section III-C2] The word 'Gaussion' should be 'Gaussian'.
  2. [Section V-D1] The word 'directely' should be 'directly'.
  3. [Fig. 11 caption] 'The frist row' should be 'The first row'.
  4. [Section IV] The sentence 'The detailed architecture of the two proposed models are are given in section IV-A and IV-B' contains a duplicated 'are'.
  5. [Equation (13)] The KLD formula has notation issues: 'P P = 1and P ˆP = 1' should read '∑ P = 1 and ∑ U P = 1' with an explicit summation index.
  6. [Section V-A] The ROC analysis is described too briefly; please specify how 'Different vs. Similar' and 'Better vs. Worse' pairs were formed, the number of pairs, and the statistical test used to produce the significance matrices.

Circularity Check

0 steps flagged · score 2.0 of 10

No derivation-level circularity: supervised models are evaluated on held-out scene splits, and the optimization claim is independently confirmed by human ratings; minor self-citations are not load-bearing.

full rationale

The central claims (BLIP2OIQA/BLIP2OISal SOTA on OHF2024) are standard supervised learning results. MOS labels (Sec. III-C1) and distortion-aware saliency maps (Sec. III-C2) are human annotations, and Tables I/III use held-out scene-based splits (Sec. V-A), so the test numbers are not fitted inputs renamed as predictions. The optimization loop in Sec. VI does use BLIP2OIQA to select low-quality images and to re-score the refined images, which is a mild self-referential element; however, Table VII also reports independent human annotation scores and shows aligned improvements (e.g., quality 46.31 to 50.66 for LQ images), so the quality-improvement claim does not reduce to the model's own score by construction. Citations [13] and [25] are from the same group and are used for motivation, perspective design, and encoder-initialization choices, but they are not invoked as a uniqueness theorem or as the sole justification for the core claims; the paper provides its own data analyses (Fig. 7, Fig. 11) and comparisons against 21 IQA and 20 saliency baselines. The unvalidated click-to-map protocol for distortion-aware saliency (no aggregation or inter-subject agreement reported) is a data-reliability risk that could affect the saliency and optimization results, but it is not a circularity between the paper's stated inputs and outputs.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claims rest on the OHF2024 database's subject labels and on the authors' annotation protocols. The free parameters are limited to a few undisclosed hyperparameters and thresholds. No new physical or conceptual entities are invented beyond the new task label 'distortion-aware saliency', which is a redefinition of existing saliency targets rather than a new entity.

free parameters (4)
  • alpha (saliency loss weight) = Not reported
    Balances the LCC and KLD terms in Eq. 11; chosen by the authors but the value is not disclosed.
  • beta (saliency loss weight) = Not reported
    Second balance weight in Eq. 11; not disclosed, making the exact loss untraceable.
  • threshold x0 in Eq. 14 = Not reported
    Binarizes the predicted saliency map into an inpainting mask; the value is set by hand and not specified.
  • viewport FOV = 110 degrees
    Set as suggested in [52]; affects the six viewport images used for feature extraction in BLIP2OIQA.
assumptions (3)
  • domain assumption MOS computed from 20 subjects per image is a reliable ground truth for the three quality dimensions
    Section III-B: all ratings come from 20 subjects aged 18-30; no inter-subject agreement, confidence intervals, or subject-level variance are reported.
  • ad hoc to paper Distortion-aware saliency maps created by smoothing click annotations with a 0.4 degree Gaussian kernel faithfully represent the regions humans find distorted
    Section III-C2: the annotation-to-map protocol is defined by the authors and is not validated against any external benchmark or inter-subject consistency check.
  • domain assumption A single 7:3 random scene-based split is sufficient to estimate model performance
    Section V-A: no repeated splits, cross-validation, or confidence intervals are reported, so the variance of the reported SRCC/PLCC values is unknown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quality Assessment and Distortion-aware Saliency Prediction for AI-Generated Omnidirectional Images." pith.science (2026). https://pith.science/paper/YQ4N6ENO

@misc{pith2026250621925,
  author       = {Pith},
  title        = {Pith review of: Quality Assessment and Distortion-aware Saliency Prediction for AI-Generated Omnidirectional Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YQ4N6ENO}},
  note         = {Machine review of arXiv:2506.21925}
}
read the original abstract

With the rapid advancement of Artificial Intelligence Generated Content (AIGC) techniques, AI generated images (AIGIs) have attracted widespread attention, among which AI generated omnidirectional images (AIGODIs) hold significant potential for Virtual Reality (VR) and Augmented Reality (AR) applications. AI generated omnidirectional images exhibit unique quality issues, however, research on the quality assessment and optimization of AI-generated omnidirectional images is still lacking. To this end, this work first studies the quality assessment and distortion-aware saliency prediction problems for AIGODIs, and further presents a corresponding optimization process. Specifically, we first establish a comprehensive database to reflect human feedback for AI-generated omnidirectionals, termed OHF2024, which includes both subjective quality ratings evaluated from three perspectives and distortion-aware salient regions. Based on the constructed OHF2024 database, we propose two models with shared encoders based on the BLIP-2 model to evaluate the human visual experience and predict distortion-aware saliency for AI-generated omnidirectional images, which are named as BLIP2OIQA and BLIP2OISal, respectively. Finally, based on the proposed models, we present an automatic optimization process that utilizes the predicted visual experience scores and distortion regions to further enhance the visual quality of an AI-generated omnidirectional image. Extensive experiments show that our BLIP2OIQA model and BLIP2OISal model achieve state-of-the-art (SOTA) results in the human visual experience evaluation task and the distortion-aware saliency prediction task for AI generated omnidirectional images, and can be effectively used in the optimization process. The database and codes will be released on https://github.com/IntMeGroup/AIGCOIQA to facilitate future research.

Figures

Figures reproduced from arXiv: 2506.21925 by the authors.

Figure 1
Figure 1. Workflow of our proposed optimization process. For the generated om [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview and subjective experiment procedure of our OHF2024 dataset. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. MOS distribution of quality, comfortability and correspondence dimensions. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Kernel distribution of four selected features of three databases: OHF2024, SUN360 [41], Matterport3D [40]. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 6
Figure 6. Figure 6: Comparison of the differences between three evaluation perspectives. (a) The left two omnidirectional images have better quality, but worse comfortability [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Illustration of the MOS correlation between any two dimensions. [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Examples of the human visual experience scores and distortion [PITH_FULL_IMAGE:figures/full_fig_p005_8.png]
Figure 9
Figure 9. Figure 9: An Overview of the proposed BLIP2OIQA model (lower part) and BLIP2OISal model (upper part). (a) Both models utilize a shared-wise encoder to [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 10
Figure 10. Figure 10: Detailed structure of the shared-wise encoder. [PITH_FULL_IMAGE:figures/full_fig_p006_10.png]
Figure 11
Figure 11. Figure 11: ROC analysis of 11 outstanding benchmark IQA models and the proposed BLIP2OIQA methods on the OHF2024 database from quality, comfortability [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]
Figure 12
Figure 12. Figure 12: Two sets of examples to illustrate the workflow and results of our optimization process. The examples include the original low-quality image and its [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

80 extracted references · 66 canonical work pages

  1. [1]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” in Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), vol. 27, 2014

    work page 2014

  2. [2]

    Auto-Encoding Variational Bayes,

    D. P. Kingma and M. Welling, “Auto-Encoding Variational Bayes,” arXiv e-prints, p. arXiv:1312.6114, Dec. 2013

    arXiv 2013

  3. [3]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10 684–10 695

    work page

  4. [4]

    Learning transferable visual models from natural language supervi- sion,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervi- sion,” in Proceedings of the International Conference on Machine Learning (ICML)

    work page

  5. [5]

    Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models,

    J. Li, D. Li, S. Savarese, and S. Hoi, “Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models,” arXiv e-prints, p. arXiv:2301.12597, Jan. 2023

    arXiv 2023

  6. [6]

    MVDiffusion: Enabling holistic multi-view image generation with correspondence- aware diffusion,

    S. Tang, F. Zhang, J. Chen, P. Wang, and Y . Furukawa, “MVDiffusion: Enabling holistic multi-view image generation with correspondence- aware diffusion,” in Proceedings of the Neural Information Processing Systems (NeurIPS), 2023

    work page 2023

  7. [7]

    Text2light: Zero-shot text-driven hdr panorama generation,

    Z. Chen, G. Wang, and Z. Liu, “Text2light: Zero-shot text-driven hdr panorama generation,” ACM Trans. Graph. , vol. 41, no. 6, nov 2022. [Online]. Available: https://doi.org/10.1145/3550454.3555447

    arXiv 2022

  8. [8]

    Dreamscene360: Unconstrained text-to- 3d scene generation with panoramic gaussian splatting,

    S. Zhou, Z. Fan, D. Xu, H. Chang, P. Chari, T. Bharadwaj, S. You, Z. Wang, and A. Kadambi, “Dreamscene360: Unconstrained text-to- 3d scene generation with panoramic gaussian splatting,” arXiv preprint arXiv:2404.06903, 2024

    arXiv 2024

Show all 80 references
  1. [9]

    4k4dgen: Panoramic 4d generation at 4k resolution,

    R. Li, P. Pan, B. Yang, D. Xu, S. Zhou, X. Zhang, Z. Li, A. Kadambi, Z. Wang, and Z. Fan, “4k4dgen: Panoramic 4d generation at 4k resolution,” arXiv preprint arXiv:2406.13527 , 2024

    2024 arXiv

  2. [10]

    GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium,” in Proceedings of the Advances in Neural Information Processing Systems (NeurIPS) , vol. 30, 2017

    2017

  3. [11]

    Improved techniques for training gans,

    T. Salimans, I. Goodfellow, W. Zaremba, V . Cheung, A. Radford, and X. Chen, “Improved techniques for training gans,” in Proceedings of the Advances in Neural Information Processing Systems (NeurIPS) , vol. 29, 2016

    2016

  4. [12]

    Clipscore: A reference-free evaluation metric for image captioning,

    J. Hessel, A. Holtzman, M. Forbes, R. Le Bras, and Y . Choi, “Clipscore: A reference-free evaluation metric for image captioning,” arXiv e-prints, p. arXiv:2104.08718, Apr. 2021

    2021 arXiv

  5. [13]

    Aigciqa2023: A large-scale image quality assessment database for ai generated images: from the perspectives of quality, authenticity and correspondence,

    J. Wang, H. Duan, J. Liu, S. Chen, X. Min, and G. Zhai, “Aigciqa2023: A large-scale image quality assessment database for ai generated images: from the perspectives of quality, authenticity and correspondence,” in Proceedings of the CAAI International Conference on Artificial ...

    2023

  6. [14]

    Agiqa-3k: An open database for ai-generated image quality assessment,

    C. Li, Z. Zhang, H. Wu, W. Sun, X. Min, X. Liu, G. Zhai, and W. Lin, “Agiqa-3k: An open database for ai-generated image quality assessment,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), 2023

    2023

  7. [15]

    Sgdnet: An end-to-end saliency-guided deep neural network for no-reference image quality assessment,

    S. Yang, Q. Jiang, W. Lin, and Y . Wang, “Sgdnet: An end-to-end saliency-guided deep neural network for no-reference image quality assessment,” in Proceedings of the ACM International Conference on Multimedia (ACM MM) , 2019, pp. 1383–1391

    2019

  8. [16]

    Realistic saliency guided image enhancement,

    S. M. H. Miangoleh, Z. Bylinskii, E. Kee, E. Shechtman, and Y . Aksoy, “Realistic saliency guided image enhancement,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 186–194

    2023

  9. [17]

    Rich human feedback for text-to-image generation,

    Y . Liang, J. He, G. Li, P. Li, A. Klimovskiy, N. Carolan, J. Sun, J. Pont- Tuset, S. Young, F. Yanget al., “Rich human feedback for text-to-image generation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2024, pp. 19 401–19 411

    2024

  10. [18]

    Mc360iqa: A multi-channel cnn for blind 360-degree image quality assessment,

    W. Sun, X. Min, G. Zhai, K. Gu, H. Duan, and S. Ma, “Mc360iqa: A multi-channel cnn for blind 360-degree image quality assessment,” IEEE Journal of Selected Topics in Signal Processing (JSTSP) , vol. 14, no. 1, pp. 64–77, 2019

    2019

  11. [19]

    Perceptual quality assessment of omnidirectional images,

    H. Duan, G. Zhai, X. Min, Y . Zhu, Y . Fang, and X. Yang, “Perceptual quality assessment of omnidirectional images,” in Proceedings of the IEEE International Symposium on Circuits and Systems (ISCAS) , 2018, pp. 1–5

    2018

  12. [20]

    Koniq-10k: An ecologically valid database for deep learning of blind image quality assessment,

    V . Hosu, H. Lin, T. Sziranyi, and D. Saupe, “Koniq-10k: An ecologically valid database for deep learning of blind image quality assessment,” IEEE Transactions on Image Processing (TIP) , vol. 29, pp. 4041–4056, 2020

    2020

  13. [21]

    Perceptual quality as- sessment of smartphone photography,

    Y . Fang, H. Zhu, Y . Zeng, K. Ma, and Z. Wang, “Perceptual quality as- sessment of smartphone photography,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (CVPR) , 2020, pp. 3677–3686

    2020

  14. [22]

    Perceptual Quality Assessment of Omnidirectional Audio- Visual Signals,

    X. Zhu, H. Duan, Y . Cao, Y . Zhu, Y . Zhu, J. Liu, L. Chen, X. Min, and G. Zhai, “Perceptual Quality Assessment of Omnidirectional Audio- Visual Signals,” in Proceedings of the CAAI International Conference on Artificial Intelligence (CICAI) , Singapore, 2024, pp. 512–525

    2024

  15. [23]

    Confusing image quality assessment: Toward better augmented reality experience,

    H. Duan, X. Min, Y . Zhu, G. Zhai, X. Yang, and P. Le Callet, “Confusing image quality assessment: Toward better augmented reality experience,” IEEE Transactions on Image Processing (TIP) , vol. 31, pp. 7206–7221, 2022

    2022

  16. [24]

    Esiqa: Perceptual quality assessment of vision-pro-based egocentric spatial images,

    X. Zhu, L. Yang, H. Duan, X. Min, G. Zhai, and P. L. Callet, “Esiqa: Perceptual quality assessment of vision-pro-based egocentric spatial images,” arXiv preprint arXiv:2407.21363 , 2024

    2024 arXiv

  17. [25]

    Understanding and evaluating human preferences for ai generated images with instruction tuning,

    J. Wang, H. Duan, G. Zhai, and X. Min, “Understanding and evaluating human preferences for ai generated images with instruction tuning,” arXiv preprint arXiv:2405.07346 , 2024

    2024 arXiv

  18. [26]

    Salicon: Saliency in context,

    M. Jiang, S. Huang, J. Duan, and Q. Zhao, “Salicon: Saliency in context,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2015, pp. 1072–1080

    2015

  19. [28]

    Cat2000: A large scale fixation dataset for boosting saliency research,

    A. Borji and L. Itti, “Cat2000: A large scale fixation dataset for boosting saliency research,” arXiv preprint arXiv:1505.03581 , 2015. 13

    2015 arXiv

  20. [29]

    Saliency based on information maximization,

    N. Bruce and J. Tsotsos, “Saliency based on information maximization,” in Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), vol. 18, 2005

    2005

  21. [30]

    Predicting human gaze using low-level saliency combined with face detection,

    M. Cerf, J. Harel, W. Einh ¨auser, and C. Koch, “Predicting human gaze using low-level saliency combined with face detection,” in Proceedings of the Advances in Neural Information Processing Systems (NeurIPS) , vol. 20, 2007

    2007

  22. [31]

    Visual saliency estimation by nonlinearly integrating features using region covariances,

    E. Erdem and A. Erdem, “Visual saliency estimation by nonlinearly integrating features using region covariances,” Journal of Vision, vol. 13, no. 4, pp. 11–11, 2013

    2013

  23. [32]

    Saliency in augmented reality,

    H. Duan, W. Shen, X. Min, D. Tu, J. Li, and G. Zhai, “Saliency in augmented reality,” in Proceedings of the ACM International Conference on Multimedia (ACM MM) , 2022

    2022

  24. [33]

    Salicon: Reducing the semantic gap in saliency prediction by adapting deep neural networks,

    X. Huang, C. Shen, X. Boix, and Q. Zhao, “Salicon: Reducing the semantic gap in saliency prediction by adapting deep neural networks,” in Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2015, pp. 262–270

    2015

  25. [34]

    Predicting human eye fixations via an lstm-based saliency attentive model,

    M. Cornia, L. Baraldi, G. Serra, and R. Cucchiara, “Predicting human eye fixations via an lstm-based saliency attentive model,” IEEE Trans- actions on Image Processing (TIP) , vol. 27, no. 10, pp. 5142–5154, 2018

    2018

  26. [35]

    Visual attention analysis and prediction on human faces for children with autism spectrum disorder,

    H. Duan, X. Min, Y . Fang, L. Fan, X. Yang, and G. Zhai, “Visual attention analysis and prediction on human faces for children with autism spectrum disorder,” ACM Transactions on Multimedia Computing, Com- munications, and Applications (TOMM), vol. 15, no. 3s, pp. 1–23, 2019

    2019

  27. [36]

    A dataset of eye movements for the children with autism spectrum disorder,

    H. Duan, G. Zhai, X. Min, Z. Che, Y . Fang, X. Yang, J. Guti ´errez, and P. L. Callet, “A dataset of eye movements for the children with autism spectrum disorder,” in Proceedings of the ACM Multimedia Systems Conference, 2019, pp. 255–260

    2019

  28. [37]

    Perceptual video quality assessment: A survey,

    X. Min, H. Duan, W. Sun, Y . Zhu, and G. Zhai, “Perceptual video quality assessment: A survey,” arXiv preprint arXiv:2402.03413 , 2024

    2024 arXiv

  29. [38]

    Blind image quality assessment: A fuzzy neural network for opinion score distribution prediction,

    Y . Gao, X. Min, Y . Zhu, X.-P. Zhang, and G. Zhai, “Blind image quality assessment: A fuzzy neural network for opinion score distribution prediction,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), 2023

    2023

  30. [39]

    Identifying children with autism spectrum disorder based on gaze-following,

    Y . Fang, H. Duan, F. Shi, X. Min, and G. Zhai, “Identifying children with autism spectrum disorder based on gaze-following,” in Proceedings of the IEEE International Conference on Image Processing (ICIP) . IEEE, 2020, pp. 423–427

    2020

  31. [40]

    Matterport3d: Learning from rgb- d data in indoor environments,

    A. Chang, A. Dai, T. Funkhouser, M. Halber, M. Niessner, M. Savva, S. Song, A. Zeng, and Y . Zhang, “Matterport3d: Learning from rgb- d data in indoor environments,” in Proceedngs of the International Conference on 3D Vision (3DV) , 2017

    2017

  32. [41]

    Recognizing scene viewpoint using panoramic place representation,

    J. Xiao, K. A. Ehinger, A. Oliva, and A. Torralba, “Recognizing scene viewpoint using panoramic place representation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2012, pp. 2695–2702

    2012

  33. [42]

    Hierarchical text-conditional image generation with clip latents,

    A. Ramesh, P. Dhariwal, A. Nichol, C. Chu, and M. Chen, “Hierarchical text-conditional image generation with clip latents,” arXiv e-prints , p. arXiv:2204.06125, Apr. 2022

    2022 arXiv

  34. [43]

    Scenescape: Text- driven consistent scene generation,

    R. Fridman, A. Abecasis, Y . Kasten, and T. Dekel, “Scenescape: Text- driven consistent scene generation,” arXiv e-prints, p. arXiv:2302.01133, Feb. 2023

    2023 arXiv

  35. [44]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR), June 2022, pp. 10 684–10 695

    2022

  36. [45]

    Audio-visual saliency for omnidirectional videos,

    Y . Zhu, X. Zhu, H. Duan, J. Li, K. Zhang, Y . Zhu, L. Chen, X. Min, and G. Zhai, “Audio-visual saliency for omnidirectional videos,” arXiv e-prints, p. arXiv:2311.05190, Nov. 2023

    2023 arXiv

  37. [46]

    Attentive deep image quality assessment for omnidirectional stitching,

    H. Duan, X. Min, W. Sun, Y . Zhu, X.-P. Zhang, and G. Zhai, “Attentive deep image quality assessment for omnidirectional stitching,” IEEE Journal of Selected Topics in Signal Processing (JSTSP) , vol. 17, no. 6, pp. 1150–1164, 2023

    2023

  38. [47]

    Augmented reality image quality assessment based on visual confusion theory,

    H. Duan, L. Guo, W. Sun, X. Min, L. Chen, and G. Zhai, “Augmented reality image quality assessment based on visual confusion theory,” in Proceedings of the IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB) , 2022, pp. 1–6

    2022

  39. [48]

    Viewing behavior supported visual saliency predictor for 360 degree videos,

    Y . Zhu, G. Zhai, Y . Yang, H. Duan, X. Min, and X. Yang, “Viewing behavior supported visual saliency predictor for 360 degree videos,” IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), vol. 32, no. 7, pp. 4188–4201, 2022

    2022

  40. [49]

    Ivqad 2017: An immersive video quality assessment database,

    H. Duan, G. Zhai, X. Yang, D. Li, and W. Zhu, “Ivqad 2017: An immersive video quality assessment database,” in Proceedings of the International Conference on Systems, Signals and Image Processing (IWSSIP), 2017, pp. 1–5

    2017

  41. [50]

    Methodology for the subjective assessment of the quality of television pictures,

    B. Series, “Methodology for the subjective assessment of the quality of television pictures,” ITU-R, Tech. Rep. ITU-R BT, 2012

    2012

  42. [51]

    How is gaze influenced by image transformations? dataset and model,

    Z. Che, A. Borji, G. Zhai, X. Min, G. Guo, and P. Le Callet, “How is gaze influenced by image transformations? dataset and model,” IEEE Transactions on Image Processing (TIP), vol. 29, pp. 2287–2300, 2019

    2019

  43. [52]

    Perceptual quality assessment of omnidirectional images as moving camera videos,

    X. Sui, K. Ma, Y . Yao, and Y . Fang, “Perceptual quality assessment of omnidirectional images as moving camera videos,” IEEE Transactions on Visualization and Computer Graphics (TVCG) , vol. 28, no. 8, pp. 3022–3034, 2021

    2021

  44. [53]

    Learning without human scores for blind image quality assessment,

    W. Xue, L. Zhang, and X. Mou, “Learning without human scores for blind image quality assessment,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE Computer Society

  45. [54]

    Blind image quality esti- mation via distortion aggravation,

    X. Min, G. Zhai, K. Gu, Y . Liu, and X. Yang, “Blind image quality esti- mation via distortion aggravation,” IEEE Transactions on Broadcasting (TBC), vol. 64, no. 2, pp. 508–517, 2018

    2018

  46. [55]

    Making a “completely blind

    A. Mittal, R. Soundararajan, and A. C. Bovik, “Making a “completely blind” image quality analyzer,” IEEE Signal Processing Letters, vol. 20, no. 3, pp. 209–212, 2013

    2013

  47. [56]

    A feature-enriched completely blind image quality evaluator,

    L. Zhang, L. Zhang, and A. C. Bovik, “A feature-enriched completely blind image quality evaluator,” IEEE Transactions on Image Processing (TIP), vol. 24, no. 8, pp. 2579–2591, 2015

    2015

  48. [57]

    Blind image quality assessment based on high order statistics aggregation,

    J. Xu, P. Ye, Q. Li, H. Du, Y . Liu, and D. S. Doermann, “Blind image quality assessment based on high order statistics aggregation,” IEEE Transactions on Image Processing (TIP), vol. 25, pp. 4444–4457, 2016. [Online]. Available: https://api.semanticscholar.org/CorpusID:1899069

    2016

  49. [58]

    Blind quality assessment based on pseudo-reference image,

    X. Min, K. Gu, G. Zhai, J. Liu, X. Yang, and C. W. Chen, “Blind quality assessment based on pseudo-reference image,” IEEE Transactions on Multimedia (TMM), vol. 20, no. 8, pp. 2049–2062, 2018

    2018

  50. [59]

    Fisblim: A five-step blind metric for quality assessment of multiply distorted images,

    K. Gu, G. Zhai, M. Liu, X. Yang, W. Zhang, X. Sun, W. Chen, and Y . Zuo, “Fisblim: A five-step blind metric for quality assessment of multiply distorted images,” in Proceedings of the SiPS , 2013, pp. 241– 246

    2013

  51. [60]

    No-reference image quality assessment in the spatial domain,

    A. Mittal, A. K. Moorthy, and A. C. Bovik, “No-reference image quality assessment in the spatial domain,” IEEE Transactions on Image Processing (TIP), vol. 21, no. 12, pp. 4695–4708, 2012

    2012

  52. [61]

    Convolutional neural networks for no-reference image quality assessment,

    L. Kang, P. Ye, Y . Li, and D. Doermann, “Convolutional neural networks for no-reference image quality assessment,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , June 2014

    2014

  53. [62]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2016, pp. 770–778

    2016

  54. [63]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv e-prints, p. arXiv:1409.1556, Sep. 2014

    2014 arXiv

  55. [64]

    Blindly assess image quality in the wild guided by a self-adaptive hyper network,

    S. Su, Q. Yan, Y . Zhu, C. Zhang, X. Ge, J. Sun, and Y . Zhang, “Blindly assess image quality in the wild guided by a self-adaptive hyper network,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2020, pp. 3664–3673

    2020

  56. [65]

    MANIQA: Multi-dimension Attention Network for No-Reference Im- age Quality Assessment,

    S. Yang, T. Wu, S. Shi, S. Lao, Y . Gong, M. Cao, J. Wang, and Y . Yang, “MANIQA: Multi-dimension Attention Network for No-Reference Im- age Quality Assessment,” arXiv e-prints , p. arXiv:2204.08958, Apr. 2022

    2022 arXiv

  57. [66]

    No-reference im- age quality assessment via transformers, relative ranking, and self- consistency,

    S. A. Golestaneh, S. Dadsetan, and K. M. Kitani, “No-reference im- age quality assessment via transformers, relative ranking, and self- consistency,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), January 2022, pp. 1220–1230

    2022

  58. [67]

    Eva: Exploring the limits of masked visual representation learning at scale,

    Y . Fang, W. Wang, B. Xie, Q. Sun, L. Wu, X. Wang, T. Huang, X. Wang, and Y . Cao, “Eva: Exploring the limits of masked visual representation learning at scale,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2023, pp. 19 358– 19 369

    2023

  59. [68]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation,

    J. Li, D. Li, C. Xiong, and S. Hoi, “Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation,” in Proceedings of the International Conference on Machine Learning (ICML), 2022, pp. 12 888–12 900

    2022

  60. [69]

    On the accuracy of objective image and video quality models: New methodology for performance evaluation,

    L. Krasula, K. Fliegel, P. Le Callet, and M. Kl ´ıma, “On the accuracy of objective image and video quality models: New methodology for performance evaluation,” in Proceedings of the International Conference on Quality of Multimedia Experience (QoMEX) . IEEE, 2016, pp. 1–6

    2016

  61. [70]

    Quality assessment of sharpened images: Challenges, methodology, and objective metrics,

    L. Krasula, P. Le Callet, K. Fliegel, and M. Kl ´ıma, “Quality assessment of sharpened images: Challenges, methodology, and objective metrics,” IEEE Transactions on Image Processing (TIP), vol. 26, no. 3, pp. 1496– 1508, 2017

    2017

  62. [71]

    Transalnet: Towards perceptually relevant visual saliency prediction,

    J. Lou, H. Lin, D. Marshall, D. Saupe, and H. Liu, “Transalnet: Towards perceptually relevant visual saliency prediction,” Neurocomputing, vol. 494, pp. 455–467, 2022. 14

    2022

  63. [72]

    Context-aware saliency detection,

    S. Goferman, L. Zelnik-Manor, and A. Tal, “Context-aware saliency detection,” IEEE Transactions on Pattern Analysis and Machine Intelli- gence (TPAMI), vol. 34, no. 10, pp. 1915–1926, 2011

    1915

  64. [73]

    Graph-based visual saliency,

    J. Harel, C. Koch, and P. Perona, “Graph-based visual saliency,” in Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), vol. 19, 2006

    2006

  65. [74]

    Visual saliency based on scale-space analysis in the frequency domain,

    J. Li, M. D. Levine, X. An, X. Xu, and H. He, “Visual saliency based on scale-space analysis in the frequency domain,” IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) , vol. 35, no. 4, pp. 996–1010, 2012

    2012

  66. [75]

    A model of saliency-based visual atten- tion for rapid scene analysis,

    L. Itti, C. Koch, and E. Niebur, “A model of saliency-based visual atten- tion for rapid scene analysis,” IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) , vol. 20, no. 11, pp. 1254–1259, 1998

    1998

  67. [76]

    Learning to predict where humans look,

    T. Judd, K. Ehinger, F. Durand, and A. Torralba, “Learning to predict where humans look,” in Proceedings of the IEEE International Confer- ence on Computer Vision (ICCV) . IEEE, 2009, pp. 2106–2113

    2009

  68. [77]

    Saliency esti- mation using a non-parametric low-level vision model,

    N. Murray, M. Vanrell, X. Otazu, and C. A. Parraga, “Saliency esti- mation using a non-parametric low-level vision model,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2011, pp. 433–440

    2011

  69. [78]

    Spatio-temporal saliency detection using phase spectrum of quaternion fourier transform,

    C. Guo, Q. Ma, and L. Zhang, “Spatio-temporal saliency detection using phase spectrum of quaternion fourier transform,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . IEEE, 2008, pp. 1–8

    2008

  70. [79]

    Saliency detection: A spectral residual approach,

    X. Hou and L. Zhang, “Saliency detection: A spectral residual approach,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Ieee, 2007, pp. 1–8

    2007

  71. [80]

    Sun: A bayesian framework for saliency using natural statistics,

    L. Zhang, M. H. Tong, T. K. Marks, H. Shan, and G. W. Cottrell, “Sun: A bayesian framework for saliency using natural statistics,” Journal of Vision, vol. 8, no. 7, pp. 32–32, 2008

    2008

  72. [81]

    Visual saliency detection by spatially weighted dissimilarity,

    L. Duan, C. Wu, J. Miao, L. Qing, and Y . Fu, “Visual saliency detection by spatially weighted dissimilarity,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . IEEE, 2011, pp. 473–480. Liu Yang received the B.E. degree from Shanghai Jia...

    2011

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.