Improved Techniques for Learning to Dehaze and Beyond: A Collective Study

Boyuan Gong; Dacheng Tao; Guanlong Zhao; Niraj Goel; Ritu Raj; Sandeep Gottimukkala; Satya Kesav; Wenqi Ren; Yang Li; Yu Liu

RECORD

Improved Techniques for Learning to Dehaze and Beyond: A Collective Study

Yu Liu, Guanlong Zhao, Boyuan Gong, Yang Li, Ritu Raj, Niraj Goel, Satya Kesav, Sandeep Gottimukkala

show 3 more authors

Zhangyang Wang, Wenqi Ren, Dacheng Tao

Computer Vision and Pattern Recognition · Submitted 30 Jun 2018

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim In both tasks, our proposed solutions significantly improve performance.

how it is stated · asserted 1.00 read the record →

Comment (0) Read on arXiv ↗

Cite ▾

arXiv ID1807.00202 (version 2)

Pith IDVDENKOQX, Pith's permanent record of this paper and its review

CitationBoyuan Gong, Dacheng Tao, Guanlong Zhao, Niraj Goel, Ritu Raj, Sandeep Gottimukkala, Satya Kesav, Wenqi Ren, Yang Li, Yu Liu, Zhangyang Wang. "Improved Techniques for Learning to Dehaze and Beyond: A Collective Study." arXiv:1807.00202 (2018). https://pith.science/paper/VDENKOQX

Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 4,248 input tokens.

What the abstract itself states, as probabilities: claims first 0.03 · states limitation 0.05 · human participants 0.03 · controlled comparison 0.23 · randomized assignment 0.01 · states quantitative result 0.04 · compares with prior results 0.34 · mentions code or data availability 0.93.

The abstract, sentence by sentence

arXiv listing abstract v2. Results and the main claim are underlined; show every role.

Here we explore two related but important tasks based on the recently released REalistic Single Image DEhazing (RESIDE) benchmark dataset: (i) single image dehazing as a low-level image restoration problem; and (ii) high-level visual understanding (e.g., object detection) of hazy images. For the first task, we investigated a variety of loss functions and show that perception-driven loss significantly improves dehazing performance. In the second task, we provide multiple solutions including using advanced modules in the dehazing-detection cascade and domain-adaptive object detectors. In both tasks, our proposed solutions significantly improve performance. GitHub repository URL is: https://github.com/guanlongzhao/dehaze

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#	sentence	role	the authors' own work	states the main conclusion	strength
s0	Here we explore two related but important tasks based on the recently released REalistic Single Image DEhazing (RESIDE) benchmark dataset: (i) single image dehazing as a low-level image restoration problem; and (ii) high-level visual understanding (e.g., object detection) of hazy images.	contribution 0.58	0.60	0.04	asserted 0.59
s1	For the first task, we investigated a variety of loss functions and show that perception-driven loss significantly improves dehazing performance.	result 0.75	0.96	0.11	asserted 1.00
s2	In the second task, we provide multiple solutions including using advanced modules in the dehazing-detection cascade and domain-adaptive object detectors.	method 0.60	0.88	0.05	asserted 0.89
s3	In both tasks, our proposed solutions significantly improve performance.	result 0.98	0.93	0.73	asserted 1.00
s4	GitHub repository URL is: https://github.com/guanlongzhao/dehaze	contribution 0.70	0.13	0.02	asserted 0.51

What kind of paper the classifier reports this is: method or tool 0.96 · evidence from observation or experiment 0.84.

What was examined, and what was not

sources read

arXiv listing abstract v2

not examined

full text
figures and tables
references and citing sentences
relations to other papers

levels run

P abstract

not run

S sections
T body sentences
C citations
R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper

Original abstract

Here we explore two related but important tasks based on the recently released REalistic Single Image DEhazing (RESIDE) benchmark dataset: (i) single image dehazing as a low-level image restoration problem; and (ii) high-level visual understanding (e.g., object detection) of hazy images. For the first task, we investigated a variety of loss functions and show that perception-driven loss significantly improves dehazing performance. In the second task, we provide multiple solutions including using advanced modules in the dehazing-detection cascade and domain-adaptive object detectors. In both tasks, our proposed solutions significantly improve performance. GitHub repository URL is: https://github.com/guanlongzhao/dehaze

Human review

How Pith checks papers → · Full audit of this paper →

References

How the literature uses it

12 arXiv papers cite this in 21 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it · 10 sentences without a confident label

Builds on it 1 paper

Inspired by Domain Adaptive Faster R-CNN in [ref] , [ref] adopts a similar idea and designs a Domain-Adaptive Mask-RCNN (DMask-RCNN).

Advanced Multiple Linear Regression Based Dark Channel Prior Applied on Dehazing Image and Generating Synthetic Haze · Domain Adaption Methods · 0.86

Compares with it 2 papers

Then the authors compare the detection accuracy with the results of dehazing-detection cascades in [ref] .

Advanced Multiple Linear Regression Based Dark Channel Prior Applied on Dehazing Image and Generating Synthetic Haze · Experiment Setup · 0.99
To evaluate the performance of FAMED-Net, we compared it with state-of-the-art image prior-based methods including DCP [ref] , FVR [ref] , BCCR [ref] , GRM [ref] , CAP [ref] , and NLD [ref] and deep CNN-based methods including DehazeNet [ref] , MSCNN [ref] , AOD-Net [ref] , FPCNet [ref] , GFN [ref] , and DCPDN [ref] .

FAMED-Net: A Fast and Accurate Multi-scale End-to-end Dehazing Network · Experiments · 1.00

Cites it as support 3 papers

The experimental results in [ref] [ref] demonstrate that the domain adaptation method can enhance the performance of both Faster R-CNN and Mask R-CNN models when tackling the object detection task in the hazy environment.

Advanced Multiple Linear Regression Based Dark Channel Prior Applied on Dehazing Image and Generating Synthetic Haze · Domain Adaption Methods · 0.99
MSCNN was chosen because recent dehazing studies [ref] endorsed it both to produce the best human-favorable, artifact-free dehazing results, and to benefit subsequent high-level task in haze most.

Single Image Deraining: A Comprehensive Benchmark Analysis · Experimental Comparison · 0.98
As advocated by many previous works [ref] , the high-level computer vision performance (such as object detection) on enhanced images could act as an indicator of the image enhancement performance itself.

All-In-One Underwater Image Enhancement using Domain-Adversarial Learning · Object detection on enhanced images · 0.97

Cites it as background 4 papers

Some object detection tasks on RESIDE RTTS dataset have been tested in [ref] [ref] [ref] .

Advanced Multiple Linear Regression Based Dark Channel Prior Applied on Dehazing Image and Generating Synthetic Haze · Experiment Setup · 0.81
These tasks encompass a wide range of practical applications, including but not limited to image super-resolution [ref] , deblurring [ref] , dehazing [ref] , inpainting [ref] , fusion [ref] , compressed sensing [ref] , low-light enhancement [ref] , and cloud removal in remote sensing [ref] .

Diffusion Models in Low-Level Vision: A Survey · Introduction · 0.85
Image enhancement as pre-processing for improving subsequent high-level vision tasks has recently received increasing attention [ref] , with a number of benchmarking efforts [ref] .

EnlightenGAN: Deep Light Enhancement without Paired Supervision · Pre-Processing for Improving Classification · 0.97

What it cites · 8 sentences without a confident label

Uses the method or tool of 1 paper

As seen from Eqn. ([ref]), the negative gradient of the domain classifier loss needs to be propagated back to ResNet, whose implementation relies on the gradient reverse layer [ref] (GRL, Fig. [ref]).

Unsupervised Domain Adaptation by Backpropagation · Solution Set 2: Domain-Adaptive Mask-RCNN · 1.00

Uses data or values from 1 paper

We benchmark against the REalistic Single Image DEhazing (RESIDE) dataset~ [ref] .

Benchmarking Single Image Dehazing and Beyond · RESIDE Dataset · 0.93

Compares with 1 paper

* the domain-adaptive detector presents a very promising approach, and its performance significantly outperforms the best results in Table [ref];[By saying that, we also emphasize that Table [ref] results have not undergone joint tuning as in [ref] , so there is potential for further improvements.]

Enhance Visual Recognition under Adverse Conditions via Deep Networks · Solution Set 2: Domain-Adaptive Mask-RCNN · 0.86

Cites as support 1 paper

Further, adding dehazing pre-processing does not always guarantee better detection (e.g, comparing RetinaNet versus AOD-Net + RetinaNet), consistent with the conclusion made in [ref] .

Benchmarking Single Image Dehazing and Beyond · Solution Set 1: Enhancing Dehazing and/or Detection Modules in the Cascade · 0.96

Cites as background 1 paper

Some of the most effective models include the multi-scale CNN (MSCNN) which predicts a coarse-scale holistic transmission map of the entire image and refines it locally [ref] ; DehazeNet, a trainable transmission matrix estimator that recovers the clean image combined with estimated global atmospheric light [ref] ; and the end-to-end dehazing network, AOD-Net [ref] , which takes a hazy image as input and directly generates a clean image output.

An All-in-One Network for Dehazing and Beyond · Haze Modeling and Dehazing Approaches · 0.93

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.