RECORD
Learning Diverse Tone Styles for Image Retouching
Computer Vision and Pattern Recognition · Submitted 12 Jul 2022
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Extensive experiments on MIT-Adobe FiveK and PPR10K datasets show that our proposed method performs favorably against state-of-the-art methods and is effective in generating diverse results to satisfy different human aesthetic preferences.
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
The abstract, sentence by sentence
Image retouching, aiming to regenerate the visually pleasing renditions of given images, is a subjective task where the users are with different aesthetic sensations. Most existing methods deploy a deterministic model to learn the retouching style from a specific expert, making it less flexible to meet diverse subjective preferences. Besides, the intrinsic diversity of an expert due to the targeted processing on different images is also deficiently described. To circumvent such issues, we propose to learn diverse image retouching with normalizing flow-based architectures. Unlike current flow-based methods which directly generate the output image, we argue that learning in a style domain could (i) disentangle the retouching styles from the image content, (ii) lead to a stable style presentation form, and (iii) avoid the spatial disharmony effects. For obtaining meaningful image tone style representations, a joint-training pipeline is delicately designed, which is composed of a style encoder, a conditional RetouchNet, and the image tone style normalizing flow (TSFlow) module. In particular, the style encoder predicts the target style representation of an input image, which serves as the conditional information in the RetouchNet for retouching, while the TSFlow maps the style representation vector into a Gaussian distribution in the forward pass. After training, the TSFlow can generate diverse image tone style vectors by sampling from the Gaussian distribution. Extensive experiments on MIT-Adobe FiveK and PPR10K datasets show that our proposed method performs favorably against state-of-the-art methods and is effective in generating diverse results to satisfy different human aesthetic preferences. Source code and pre-trained models are publicly available at https://github.com/SSRHeart/TSFlow.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Image retouching, aiming to regenerate the visually pleasing renditions of given images, is a subjective task where the users are with different aesthetic sensations. | background 1.00 | 0.05 | 0.02 | asserted 1.00 |
| s1 | Most existing methods deploy a deterministic model to learn the retouching style from a specific expert, making it less flexible to meet diverse subjective preferences. | gap 0.70 | 0.04 | 0.03 | asserted 1.00 |
| s2 | Besides, the intrinsic diversity of an expert due to the targeted processing on different images is also deficiently described. | gap 0.97 | 0.05 | 0.02 | asserted 1.00 |
| s3 | To circumvent such issues, we propose to learn diverse image retouching with normalizing flow-based architectures. | contribution 0.98 | 0.60 | 0.08 | asserted 0.89 |
| s4 | Unlike current flow-based methods which directly generate the output image, we argue that learning in a style domain could (i) disentangle the retouching styles from the image content, (ii) lead to a stable style presentation form, and (iii) avoid the spatial disharmony effects. | contribution 0.61 | 0.22 | 0.11 | hedged possibility 0.80 |
| s5 | For obtaining meaningful image tone style representations, a joint-training pipeline is delicately designed, which is composed of a style encoder, a conditional RetouchNet, and the image tone style normalizing flow (TSFlow) module. | method 0.99 | 0.91 | 0.04 | asserted 1.00 |
| s6 | In particular, the style encoder predicts the target style representation of an input image, which serves as the conditional information in the RetouchNet for retouching, while the TSFlow maps the style representation vector into a Gaussian distribution in the forward pass. | method 1.00 | 0.85 | 0.03 | asserted 1.00 |
| s7 | After training, the TSFlow can generate diverse image tone style vectors by sampling from the Gaussian distribution. | method 0.99 | 0.83 | 0.09 | asserted 1.00 |
| s8 | Extensive experiments on MIT-Adobe FiveK and PPR10K datasets show that our proposed method performs favorably against state-of-the-art methods and is effective in generating diverse results to satisfy different human aesthetic preferences. | result 1.00 | 0.96 | 0.78 | asserted 1.00 |
| s9 | Source code and pre-trained models are publicly available at https://github.com/SSRHeart/TSFlow. | contribution 0.49 | 0.45 | 0.02 | asserted 0.98 |
What was examined, and what was not
- arXiv listing abstract v2
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Image retouching, aiming to regenerate the visually pleasing renditions of given images, is a subjective task where the users are with different aesthetic sensations. Most existing methods deploy a deterministic model to learn the retouching style from a specific expert, making it less flexible to meet diverse subjective preferences. Besides, the intrinsic diversity of an expert due to the targeted processing on different images is also deficiently described. To circumvent such issues, we propose to learn diverse image retouching with normalizing flow-based architectures. Unlike current flow-based methods which directly generate the output image, we argue that learning in a style domain could (i) disentangle the retouching styles from the image content, (ii) lead to a stable style presentation form, and (iii) avoid the spatial disharmony effects. For obtaining meaningful image tone style representations, a joint-training pipeline is delicately designed, which is composed of a style encoder, a conditional RetouchNet, and the image tone style normalizing flow (TSFlow) module. In particular, the style encoder predicts the target style representation of an input image, which serves as the conditional information in the RetouchNet for retouching, while the TSFlow maps the style representation vector into a Gaussian distribution in the forward pass. After training, the TSFlow can generate diverse image tone style vectors by sampling from the Gaussian distribution. Extensive experiments on MIT-Adobe FiveK and PPR10K datasets show that our proposed method performs favorably against state-of-the-art methods and is effective in generating diverse results to satisfy different human aesthetic preferences. Source code and pre-trained models are publicly available at https://github.com/SSRHeart/TSFlow.
Human review
How the literature uses it
2 arXiv papers cite this in 5 sentences
Papers citing it · 4 sentences without a confident label
Cites it as background 1 paper
-
Conditional normalizing flows [ref] have been explored for various inverse problems in imaging such as super-resolution space generation [ref] , low-light image enhancement [ref] , guided image generation [ref] , image dehazing [ref] , denoising [ref] and inpainting [ref] .
On the Robustness of Normalizing Flows for Inverse Problems in Imaging · Introduction · 0.88
What it cites · 12 sentences without a confident label
Contests 1 paper
-
LLFlow [ref] (the middle column of <ref>) provides more information by taking the predicted color map $𝐂$ as the mean of the Gaussian distribution, , $𝐙∼𝒩(𝐂, 𝐈)$, yet the spatial disharmony problem still remains unsolved.
Low-Light Image Enhancement with Normalizing Flow · Motivation · contest 0.80
Uses the method or tool of 7 papers
-
Following previous conditional normalizing flow-based methods [ref] , our TSFlow is comprised of a sequence of flow steps, which contains a conditional coupling block [ref] , an invertible $1×1$ convolution, and an actnorm layer.
Low-Light Image Enhancement with Normalizing Flow · Network Structure · 0.91
-
Following previous conditional normalizing flow-based methods [ref] , our TSFlow is comprised of a sequence of flow steps, which contains a conditional coupling block [ref] , an invertible $1×1$ convolution, and an actnorm layer.
Guided Image Generation with Conditional Invertible Neural Networks · Network Structure · 0.91
-
With the normalizing flow architecture [ref] , we are able to describe the probability distribution of the image tone style.
Density estimation using Real NVP · Learning Objective · 0.93
Cites as background 4 papers
-
Normalizing flow [ref] is a powerful generative model with invertible structures, which maps a complex distribution into a simple one (, Gaussian distribution) in the forward pass, and achieves diverse generations by sampling from the simple distribution in the reverse process.
NICE: Non-linear Independent Components Estimation · Conditional Normalizing Flow · 0.99
-
Normalizing flow [ref] is a powerful generative model with invertible structures, which maps a complex distribution into a simple one (, Gaussian distribution) in the forward pass, and achieves diverse generations by sampling from the simple distribution in the reverse process.
Density estimation using Real NVP · Conditional Normalizing Flow · 0.99
-
Normalizing flow [ref] is a powerful generative model with invertible structures, which maps a complex distribution into a simple one (, Gaussian distribution) in the forward pass, and achieves diverse generations by sampling from the simple distribution in the reverse process.
Glow: Generative Flow with Invertible 1x1 Convolutions · Conditional Normalizing Flow · 0.99
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.