arxiv: 2604.16284 · v1 · submitted 2026-04-17 · 💻 cs.CV

Recognition: unknown

Enhancing Hazy Wildlife Imagery: AnimalHaze3k and IncepDehazeGan

Shivarth Rai , Tejeswar Pokuri

Authors on Pith no claims yet

Pith reviewed 2026-05-10 08:14 UTC · model grok-4.3

classification 💻 cs.CV

keywords image dehazingwildlife imagerygenerative adversarial networkssynthetic datasetanimal detectioncomputer visionimage restorationconservation monitoring

0 comments

The pith

A new GAN architecture trained on synthetic hazy wildlife photos restores image clarity enough to more than double object detection accuracy.

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

The paper creates AnimalHaze3k, a collection of 3,477 synthetic hazy images derived from 1,159 clear wildlife photographs via a physics-based haze simulation. It introduces IncepDehazeGan, which places inception blocks and residual skip connections inside a generative adversarial network to remove atmospheric haze. The model records SSIM of 0.8914, PSNR of 20.54, and LPIPS of 0.1104, exceeding prior dehazing techniques. When the restored images feed into YOLOv11, mean average precision rises by 112 percent and intersection-over-union by 67 percent. These steps are presented as practical support for ecologists monitoring animals under reduced visibility.

Core claim

We introduce AnimalHaze3k, a synthetic dataset of 3,477 hazy wildlife images generated from 1,159 clear photographs through a physics-based pipeline, and IncepDehazeGan, a GAN that fuses inception blocks with residual skip connections, which achieves SSIM 0.8914, PSNR 20.54, and LPIPS 0.1104 while lifting YOLOv11 mAP by 112 percent and IoU by 67 percent on the dehazed outputs.

What carries the argument

IncepDehazeGan, a generative adversarial network that inserts inception blocks for multi-scale feature capture and residual skip connections for stable gradient flow during haze removal from wildlife photographs.

If this is right

Dehazed wildlife images produce substantially higher accuracy in automated animal detection and tracking tasks.
The synthetic AnimalHaze3k dataset supplies training data for dehazing models when real paired hazy-clear wildlife images are scarce.
Improved visibility in restored frames supports downstream conservation tasks such as population counts and behavior studies.
The reported metric gains establish a performance baseline for specialized haze removal in narrow-domain imagery.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the synthetic haze model generalizes, the same architecture could be retrained on hazy images from other domains such as aerial surveys or underwater footage.
A direct side-by-side test on real hazy field data would reveal whether the reported detection gains hold outside the synthetic distribution.
Coupling the dehazer with tracking algorithms could reduce identity switches caused by haze-induced appearance changes.

Load-bearing premise

The physics-based pipeline that adds synthetic haze to clear wildlife photographs produces images that match the visual and statistical properties of real atmospheric haze encountered in the field.

What would settle it

Run IncepDehazeGan on a held-out collection of actual field-captured hazy wildlife photographs, measure the resulting SSIM/PSNR/LPIPS against human-labeled clear references, and compare the YOLOv11 mAP and IoU gains to the numbers obtained on the synthetic test set.

Figures

Figures reproduced from arXiv: 2604.16284 by Shivarth Rai, Tejeswar Pokuri.

**Figure 3.** Figure 3: IncepDehazeGan Generator architecture [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: IncepDehazeGan Discriminator architecture [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

**Figure 5.** Figure 5: Qualitative comparison on dehazing results across SOTA models and IncepDehazeGan (our model). [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗

**Figure 6.** Figure 6: YOLOv11 animal detection on (a,d) ground truth, (b,e) [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗

read the original abstract

Atmospheric haze significantly degrades wildlife imagery, impeding computer vision applications critical for conservation, such as animal detection, tracking, and behavior analysis. To address this challenge, we introduce AnimalHaze3k a synthetic dataset comprising of 3,477 hazy images generated from 1,159 clear wildlife photographs through a physics-based pipeline. Our novel IncepDehazeGan architecture combines inception blocks with residual skip connections in a GAN framework, achieving state-of-the-art performance (SSIM: 0.8914, PSNR: 20.54, and LPIPS: 0.1104), delivering 6.27% higher SSIM and 10.2% better PSNR than competing approaches. When applied to downstream detection tasks, dehazed images improved YOLOv11 detection mAP by 112% and IoU by 67%. These advances can provide ecologists with reliable tools for population monitoring and surveillance in challenging environmental conditions, demonstrating significant potential for enhancing wildlife conservation efforts through robust visual analytics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper's real contribution is a new synthetic wildlife haze dataset, but its SOTA claims and downstream gains rest only on that synthetic data with no real hazy field tests or baseline details.

read the letter

The one thing to know is that this work creates AnimalHaze3k, a set of roughly 3500 synthetic hazy wildlife images generated from clear photos via a standard physics-based scattering model, and pairs it with IncepDehazeGan, which is a GAN that adds inception blocks and residual skips to a fairly conventional dehazing setup. That dataset is the clearest new element; the architecture itself is a domain-specific recombination rather than a novel mechanism. The paper does a reasonable job of showing that dehazing the synthetic images improves YOLOv11 detection numbers on the same synthetic test split, and the reported SSIM, PSNR, and LPIPS values are competitive within the numbers they give. For someone building tools for hazy-condition animal monitoring, the dataset could be a practical starting point if it is released with the generation code. The soft spots are straightforward and fairly large. Everything is evaluated only on the synthetic data; the abstract and stress-test note give no quantitative or qualitative results on actual hazy field photographs, so the generalization claim to real conservation use is untested. The 112% mAP and 67% IoU gains are presented without the raw baseline mAP or IoU values, without any ablation on the inception or residual components, and without details on how the percentages were computed. That makes it difficult to judge whether the improvement is as large as stated or simply reflects a weak starting point. The physics pipeline is standard, but the paper does not appear to include any cross-check against real atmospheric measurements or real hazy wildlife scenes. This paper is mainly for applied computer-vision groups working on ecology or remote sensing who need domain-specific hazy data. A reader looking for a ready-to-use wildlife dehazing benchmark might find the dataset useful once released. It deserves a serious referee because the dataset is new and the downstream task is relevant, but the review would need to require real-data validation and clearer experimental reporting before acceptance. I would send it to review with those specific requests rather than desk-reject.

Referee Report

3 major / 2 minor

Summary. The paper introduces the AnimalHaze3k synthetic dataset of 3,477 hazy wildlife images generated from 1,159 clear photographs via a physics-based atmospheric scattering pipeline. It proposes IncepDehazeGan, a GAN that integrates inception blocks with residual skip connections, and reports state-of-the-art dehazing metrics on this dataset (SSIM 0.8914, PSNR 20.54, LPIPS 0.1104) along with 6.27% SSIM and 10.2% PSNR gains over prior methods. The work further claims that applying the dehazer to YOLOv11 detection yields 112% higher mAP and 67% higher IoU, positioning the contributions as aids for wildlife conservation in hazy conditions.

Significance. If the synthetic haze model generalizes, the domain-specific dataset and tailored GAN architecture would constitute a useful contribution to hazy-image enhancement for wildlife monitoring, potentially supporting downstream tasks like animal detection in conservation settings. The explicit focus on a wildlife domain and the reported downstream detection gains distinguish the work from generic dehazing papers, though the absence of real-field validation constrains the assessed impact.

major comments (3)

[Abstract and Results] Abstract and Results section: the claimed 112% mAP and 67% IoU improvements for YOLOv11 are presented without baseline mAP or IoU values on the original hazy images and without any description of how the percentages were computed, rendering the magnitude of the downstream benefit unverifiable.
[Dataset generation and Experiments] Dataset generation and Experiments sections: all quantitative results (SSIM, PSNR, LPIPS, and detection metrics) are obtained exclusively on the synthetic AnimalHaze3k data; no quantitative or qualitative evaluation on authentic hazy field imagery is reported, leaving the central claim of utility for real-world wildlife conservation untested.
[Methodology] Methodology section: the paper introduces IncepDehazeGan as a novel combination of inception blocks and residual skip connections inside a GAN but provides no ablation studies isolating the contribution of each component, so the architectural rationale for the reported gains cannot be assessed.

minor comments (2)

[Abstract] The abstract asserts 'state-of-the-art' performance but does not tabulate the exact scores of the competing methods being compared.
[Dataset] Training/validation/test split sizes and the precise atmospheric scattering parameters used in the physics-based pipeline are not stated in the provided summary.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment point by point below, indicating the revisions we will incorporate where feasible.

read point-by-point responses

Referee: [Abstract and Results] Abstract and Results section: the claimed 112% mAP and 67% IoU improvements for YOLOv11 are presented without baseline mAP or IoU values on the original hazy images and without any description of how the percentages were computed, rendering the magnitude of the downstream benefit unverifiable.

Authors: We agree that the baseline mAP and IoU values on the original hazy images, along with the exact computation method, should have been provided to make the reported gains verifiable. In the revised manuscript, we will include these baseline values for YOLOv11 on the hazy images and explicitly describe the percentage improvement formula as ((dehazed_metric - hazy_metric) / hazy_metric) * 100%. revision: yes
Referee: [Dataset generation and Experiments] Dataset generation and Experiments sections: all quantitative results (SSIM, PSNR, LPIPS, and detection metrics) are obtained exclusively on the synthetic AnimalHaze3k data; no quantitative or qualitative evaluation on authentic hazy field imagery is reported, leaving the central claim of utility for real-world wildlife conservation untested.

Authors: Our study is designed around the synthetic AnimalHaze3k dataset to enable precise quantitative evaluation using known ground-truth clear images. We acknowledge the value of real-world validation but note that paired real hazy and clear wildlife images with accurate annotations are not available to us. revision: partial
Referee: [Methodology] Methodology section: the paper introduces IncepDehazeGan as a novel combination of inception blocks and residual skip connections inside a GAN but provides no ablation studies isolating the contribution of each component, so the architectural rationale for the reported gains cannot be assessed.

Authors: We recognize that ablation studies are necessary to isolate the contributions of the inception blocks and residual skip connections. In the revised manuscript, we will add ablation experiments that systematically remove each component and report the resulting changes in SSIM, PSNR, and LPIPS on the AnimalHaze3k test set. revision: yes

standing simulated objections not resolved

We cannot provide quantitative or qualitative evaluation on authentic hazy field imagery because no such paired real-world dataset with ground truth is available.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces a new synthetic dataset (AnimalHaze3k) generated from clear wildlife photographs via a described physics-based pipeline and presents a novel GAN architecture (IncepDehazeGan) whose performance is evaluated using standard metrics (SSIM, PSNR, LPIPS) plus downstream detection on the same synthetic data. No equations, derivations, or self-referential definitions are provided that would reduce the reported gains to quantities defined by fitted parameters within the paper itself; the results are empirical outputs from training and testing rather than predictions forced by construction. Any self-citations are not load-bearing for the central claims, and the work remains self-contained against external benchmarks without invoking uniqueness theorems or ansatzes from prior author work.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claims rest on the assumption that a standard atmospheric scattering model produces representative hazy wildlife images and that the GAN trained on those images will perform well on real haze.

free parameters (1)

Atmospheric scattering parameters
Values for transmission map and atmospheric light chosen to synthesize hazy versions of the clear images.

axioms (1)

domain assumption The atmospheric scattering model accurately represents haze in natural wildlife scenes
Invoked to generate the synthetic AnimalHaze3k dataset from clear photographs.

pith-pipeline@v0.9.0 · 5480 in / 1532 out tokens · 57100 ms · 2026-05-10T08:14:42.088414+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

39 extracted references · 7 canonical work pages · 1 internal anchor

[1]

ZoeDepth: Zero-shot Transfer by Combining Relative and Metric Depth

Shariq Farooq Bhat, Reiner Birkl, Diana Wofk, Peter Wonka, and Matthias M ¨uller. Zoedepth: Zero-shot trans- fer by combining relative and metric depth.arXiv preprint arXiv:2302.12288, 2023. 2

work page internal anchor Pith review arXiv 2023
[2]

Dehazenet: An end-to-end system for single image haze removal.IEEE transactions on image process- ing, 25(11):5187–5198, 2016

Bolun Cai, Xiangmin Xu, Kui Jia, Chunmei Qing, and Dacheng Tao. Dehazenet: An end-to-end system for single image haze removal.IEEE transactions on image process- ing, 25(11):5187–5198, 2016. 1

2016
[3]

Conservation ai: Live stream analysis for the detection of endangered species using convolutional neural networks and drone technology.arXiv preprint arXiv:1910.07360, 2019

Carl Chalmers, Paul Fergus, Serge Wich, and Aday Curbelo Montanez. Conservation ai: Live stream analysis for the detection of endangered species using convolutional neural networks and drone technology.arXiv preprint arXiv:1910.07360, 2019. 1

work page arXiv 1910
[4]

Dea-net: Single image dehazing based on detail-enhanced convolution and content-guided attention, 2023

Zixuan Chen, Zewei He, and Zhe-Ming Lu. Dea-net: Single image dehazing based on detail-enhanced convolution and content-guided attention, 2023. 3

2023
[5]

Multi-scale boosted de- hazing network with dense feature fusion

Hang Dong, Jinshan Pan, Lei Xiang, Zhe Hu, Xinyi Zhang, Fei Wang, and Ming-Hsuan Yang. Multi-scale boosted de- hazing network with dense feature fusion. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2157–2167, 2020. 1

2020
[6]

Physics-based feature de- hazing networks

Jiangxin Dong and Jinshan Pan. Physics-based feature de- hazing networks. InComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXX 16, pages 188–204. Springer, 2020. 1

2020
[7]

FD-GAN: generative adversarial networks with fusion-discriminator for single image dehazing.CoRR, abs/2001.06968, 2020

Yu Dong, Yihao Liu, He Zhang, Shifeng Chen, and Yu Qiao. FD-GAN: generative adversarial networks with fusion-discriminator for single image dehazing.CoRR, abs/2001.06968, 2020. 3

work page arXiv 2001
[8]

Dehazing network: Asymmetric unet based on physical model.IEEE Transactions on Geoscience and Remote Sensing, 62:1–12, 2024

Yang Du, Jun Li, Qinghong Sheng, Yuxin Zhu, Bo Wang, and Xiao Ling. Dehazing network: Asymmetric unet based on physical model.IEEE Transactions on Geoscience and Remote Sensing, 62:1–12, 2024. 3

2024
[9]

Hybriddepth: Ro- bust metric depth fusion by leveraging depth from focus and single-image priors.arXiv preprint arXiv:2407.18443, 2024

Ashkan Ganj, Hang Su, and Tian Guo. Hybriddepth: Ro- bust metric depth fusion by leveraging depth from focus and single-image priors.arXiv preprint arXiv:2407.18443, 2024. 2

work page arXiv 2024
[10]

Face recognition of a lorisidae species based on computer vision.Global Ecology and Conservation, 45: e02511, 2023

Yan Guan, Yujie Lei, Yuhui Zhu, Tingxuan Li, Ying Xiang, Pengmei Dong, Rong Jiang, Jinwen Luo, Anqi Huang, Yu- mai Fan, et al. Face recognition of a lorisidae species based on computer vision.Global Ecology and Conservation, 45: e02511, 2023. 1

2023
[11]

Alain Hor ´e and Djemel Ziou. Image quality metrics: Psnr (a) Hazy (b) FFA (c) FD-GAN (d) DEANET (e) DehazeFormer (f) IncepDehaze (g) Ground Truth (h) Hazy (i) FFA (j) FD-GAN (k) DEANET (l) DehazeFormer (m) IncepDehaze (n) Ground Truth Figure 5. Qualitative comparison on dehazing results across SOTA models and IncepDehazeGan (our model). (a) (b) (c) (d) (...

2010
[12]

Identification of animal individuals using deep learning: A case study of giant panda.Biological Conservation, 242:108414, 2020

Jin Hou, Yuxin He, Hongbo Yang, Thomas Connor, Jie Gao, Yujun Wang, Yichao Zeng, Jindong Zhang, Jinyan Huang, Bochuan Zheng, et al. Identification of animal individuals using deep learning: A case study of giant panda.Biological Conservation, 242:108414, 2020. 1

2020
[13]

Image-to-image translation with conditional adver- sarial networks

Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adver- sarial networks. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1125–1134,
[14]

Yolov11: An overview of the key architectural enhancements, 2024

Rahima Khanam and Muhammad Hussain. Yolov11: An overview of the key architectural enhancements, 2024. 3

2024
[15]

Benchmarking single- image dehazing and beyond.IEEE Transactions on Image Processing, 28(1):492–505, 2018

Boyi Li, Wenqi Ren, Dengpan Fu, Dacheng Tao, Dan Feng, Wenjun Zeng, and Zhangyang Wang. Benchmarking single- image dehazing and beyond.IEEE Transactions on Image Processing, 28(1):492–505, 2018. 2

2018
[16]

Multi-layers feature fusion of convolutional neural network for scene classification of remote sensing.IEEE Access, 7:121685– 121694, 2019

Chenhui Ma, Xiaodong Mu, and Dexuan Sha. Multi-layers feature fusion of convolutional neural network for scene classification of remote sensing.IEEE Access, 7:121685– 121694, 2019. 3

2019
[17]

Vision and the atmosphere.International journal of computer vision, 48: 233–254, 2002

Srinivasa G Narasimhan and Shree K Nayar. Vision and the atmosphere.International journal of computer vision, 48: 233–254, 2002. 1

2002
[18]

Interactive (de) weathering of an image using physical models

Srinivasa G Narasimhan and Shree K Nayar. Interactive (de) weathering of an image using physical models. InIEEE Workshop on color and photometric Methods in computer Vision, page 1. France, 2003. 1

2003
[19]

Limitations of recreational camera traps for wildlife manage- ment and conservation research: A practitioner’s perspec- tive.Ambio, 44:624–635, 2015

Scott Newey, Paul Davidson, Sajid Nazir, Gorry Fairhurst, Fabio Verdicchio, R Justin Irvine, and Ren ´e Van Der Wal. Limitations of recreational camera traps for wildlife manage- ment and conservation research: A practitioner’s perspec- tive.Ambio, 44:624–635, 2015. 1

2015
[20]

Mohammad Sadegh Norouzzadeh, Anh Nguyen, Margaret Kosmala, Alexandra Swanson, Meredith S Palmer, Craig Packer, and Jeff Clune. Automatically identifying, count- ing, and describing wild animals in camera-trap images with deep learning.Proceedings of the National Academy of Sci- ences, 115(25):E5716–E5725, 2018. 1

2018
[21]

Camera traps in animal ecology: methods and analyses

Allan F O’Connell, James D Nichols, and K Ullas Karanth. Camera traps in animal ecology: methods and analyses. Springer, 2011. 1

2011
[22]

Ffa-net: Feature fusion attention network for single image dehazing.CoRR, abs/1911.07559, 2019

Xu Qin, Zhilin Wang, Yuanchao Bai, Xiaodong Xie, and Huizhu Jia. Ffa-net: Feature fusion attention network for single image dehazing.CoRR, abs/1911.07559, 2019. 3

work page arXiv 1911
[23]

Tracking and analysing social interactions in dairy cattle with real-time locating system and machine learning.Jour- nal of Systems Architecture, 116:102139, 2021

Keni Ren, Gun Bernes, M ˚arten Hetta, and Johannes Karls- son. Tracking and analysing social interactions in dairy cattle with real-time locating system and machine learning.Jour- nal of Systems Architecture, 116:102139, 2021. 1

2021
[24]

Single image dehazing via multi- scale convolutional neural networks

Wenqi Ren, Si Liu, Hua Zhang, Jinshan Pan, Xiaochun Cao, and Ming-Hsuan Yang. Single image dehazing via multi- scale convolutional neural networks. InComputer Vision– ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14, pages 154–169. Springer, 2016. 1

2016
[25]

Hubot: A biomimicking mobile robot for non- disruptive bird behavior study.Ecological Informatics, 85: 102939, 2025

Lyes Saad Saoud, Lo ¨ıc Lesobre, Enrico Sorato, Saud Al Qaydi, Yves Hingrat, Lakmal Seneviratne, and Irfan Hussain. Hubot: A biomimicking mobile robot for non- disruptive bird behavior study.Ecological Informatics, 85: 102939, 2025. 1

2025
[26]

Identification of an- imals and recognition of their actions in wildlife videos us- ing deep learning techniques.Ecological Informatics, 61: 101215, 2021

Frank Schindler and V olker Steinhage. Identification of an- imals and recognition of their actions in wildlife videos us- ing deep learning techniques.Ecological Informatics, 61: 101215, 2021. 1

2021
[27]

A comprehensive review of computational dehazing techniques.Archives of Computa- tional Methods in Engineering, 26(5):1395–1413, 2019

Dilbag Singh and Vijay Kumar. A comprehensive review of computational dehazing techniques.Archives of Computa- tional Methods in Engineering, 26(5):1395–1413, 2019. 1

2019
[28]

Vision transformers for single image dehazing.IEEE Transactions on Image Processing, 32:1927–1941, 2023

Yuda Song, Zhuqing He, Hui Qian, and Xin Du. Vision transformers for single image dehazing.IEEE Transactions on Image Processing, 32:1927–1941, 2023. 2, 3

1927
[29]

Going Deeper with Convolutions

Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott E. Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions.CoRR, abs/1409.4842, 2014. 3

work page Pith review arXiv 2014
[30]

An- imal detection and classification from camera trap images us- ing different mainstream object detection architectures.Ani- mals, 12(15):1976, 2022

Mengyu Tan, Wentao Chao, Jo-Ku Cheng, Mo Zhou, Yiwen Ma, Xinyi Jiang, Jianping Ge, Lian Yu, and Limin Feng. An- imal detection and classification from camera trap images us- ing different mainstream object detection architectures.Ani- mals, 12(15):1976, 2022. 2

1976
[31]

Attention is all you need.Advances in neural information processing systems, 30, 2017

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017. 2

2017
[32]

Recent advances in im- age dehazing.IEEE/CAA Journal of Automatica Sinica, 4 (3):410–436, 2017

Wencheng Wang and Xiaohui Yuan. Recent advances in im- age dehazing.IEEE/CAA Journal of Automatica Sinica, 4 (3):410–436, 2017. 1

2017
[33]

Bovik, H.R

Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing, 13(4): 600–612, 2004. 3

2004
[34]

Con- trastive learning for compact single image dehazing

Haiyan Wu, Yanyun Qu, Shaohui Lin, Jian Zhou, Ruizhi Qiao, Zhizhong Zhang, Yuan Xie, and Lizhuang Ma. Con- trastive learning for compact single image dehazing. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10551–10560, 2021. 1

2021
[35]

Deep depth from focus with differential focus volume

Fengting Yang, Xiaolei Huang, and Zihan Zhou. Deep depth from focus with differential focus volume. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12642–12651, 2022. 2

2022
[36]

Depth anything: Unleashing the power of large-scale unlabeled data

Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10371–10381, 2024. 2

2024
[37]

Densely connected pyramid dehazing network

He Zhang and Vishal M Patel. Densely connected pyramid dehazing network. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3194–3203,
[38]

Fsim: A feature similarity index for image quality assess- ment.IEEE Transactions on Image Processing, 20(8):2378– 2386, 2011

Lin Zhang, Lei Zhang, Xuanqin Mou, and David Zhang. Fsim: A feature similarity index for image quality assess- ment.IEEE Transactions on Image Processing, 20(8):2378– 2386, 2011. 3

2011
[39]

The Unreasonable Effectiveness of Deep Features as a Perceptual Metric

Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric.CoRR, abs/1801.03924,

work page Pith review arXiv