A Controlled Study on Machine Learning Applications to Predict Dry Fabric Color from Wet Samples: Influences of Dye Concentration and Squeeze Pressure
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsPg 5
Specify DI water
Explain line 182
Explain figure 3 in the main text
Pg 6
Line 1: why there are 6 concentrations?
Pg 7
Plz provide a breakdown of 762 samples. Possibly using a table.
Is it one sample dyeing per machine run?
Line 239: “About 37% of the samples were dyed with a single primary dye, 40% of the samples were dyed with two-dye combinations, and the remaining 23% of the samples were dyed with three-dye combinations.” – need to include the dye ratios of these combinations
Pg 9
Table S4, S5, and S6 are missing
Pg 11-12
Plz give some explanation on how “Baseline model for wet and dry states color difference works for comparison to actual and predicted dry L*a*b* values prediction error.” For figure 7a-e. Does this baseline model based on delta-E2000 > 1.0? If yes, then add this explanation.
Need to provide each of the five models output/model/test parameters in a table. Regression model’s output needs a separate table.
Figure 3 says, pressure, temperature and speed were inputs to the neural network model, but it looks like they are not used in actual calculations.
Page 14
Line 436: “…area is devoid of data points due to the fact these dyes are not brighteners.” – needs explanation of this part.
Line 436: “This distribution reflects the impact of the dyeing process, which can be influenced by factors such as fluctuations in dye concentrations, subtle variations in squeeze roller pressure, and the inherent variability of the textile fabric.” -also the dye combination is responsible.
“The baseline model consistently produced high CIEDE2000 error values, highlighting the inability of non-machine learning approaches to capture the complex relationships between wet and dry color states.” – so, baseline model capture only the differences between dry and wet state, and it could vary a lot due to roller pressure, given that fabric physical properties are same. So, there are valid reasons for having Delta > 1. Then why is it considered as an error? Dry state is the target state that the machine learning model predicts.
Page 15: figure and table number needs correction
Line: 550: the table says it is 63.9%. not above 64%.
Author Response
Reviewer 1 Comments
Open Review
(x) I would not like to sign my review report
( ) I would like to sign my review report
Quality of English Language
( ) The English could be improved to more clearly express the research.
(x) The English is fine and does not require any improvement.
|
|
Yes |
Can be improved |
Must be improved |
Not applicable |
Does the introduction provide sufficient background and include all relevant references? |
(x) |
( ) |
( ) |
( ) |
Is the research design appropriate? |
(x) |
( ) |
( ) |
( ) |
Are the methods adequately described? |
(x) |
( ) |
( ) |
( ) |
Are the results clearly presented? |
( ) |
(x) |
( ) |
( ) |
Are the conclusions supported by the results? |
(x) |
( ) |
( ) |
( ) |
Comments and Suggestions for Authors
Thank you for your feedback.
Pg 5
Specify DI water
Answer:
The abbreviation “DI” was added to the first occurrence of deionized water. In subsequent occurrences “deionized water” was replaced with “DI water”.
Explain line 182
Answer:
Line#182:
“Since CIELAB is not a perceptually uniform color space, the magnitude of the distance between any two colors (points in three space) does not always equate with the visu-ally perceived color difference. This should be considered when evaluating the model performance as presented in this study.”
For clarity this section was reworded to:
Typically, machine learning models use the Euclidean norm as the error function for model training. However, since color space is not perceptually uniform (ie Euclidean), the Euclidean norm is not the best metric to describe the color distance between points. This should be considered when training models and evaluating model performance.
Explain figure 3 in the main text
Answer:
The Figure 3 was replaced to explain the text in section “2.2 Fabric Dyeing Procedure”.
Pg 6
Line 1: why there are 6 concentrations?
Answer:
Models were trained using a combination of dyes at six dye concentrations (0.1%, 0.2%, 0.25%, 0.5%, 1%, and 2%) which covered the color gamut for these dyes. This allows the model to predict the color of a dried fabric when dyed with any of these six concentrations. Dye concentrations <0.1% and >2% do not affect the ultimate shade with these dyes.
Pg 7
Plz provide a breakdown of 762 samples. Possibly using a table.
Answer:
This information is in the manuscript in section “2.5 Color Data”
Is it one sample dyeing per machine run?
For our sample batch dyeing machine, we can dye 15 samples simultaneously per run.
Line 239: “About 37% of the samples were dyed with a single primary dye, 40% of the samples were dyed with two-dye combinations, and the remaining 23% of the samples were dyed with three-dye combinations.” – need to include the dye ratios of these combinations
Answer:
We appreciate the suggestion. However, the primary focus of this study is on the color transformation from wet to dry states rather than the specifics of the dyeing process itself. While the dye ratios were controlled and recorded, they are not directly relevant to the machine learning models, which rely on color measurements rather than formulation details. That said, for transparency, this data is available upon request, as noted in the ‘Data Availability Statement.’
Pg 9
Table S4, S5, and S6 are missing
Answer: These tables are included in the supplementary section.
Pg 11-12
Plz give some explanation on how “Baseline model for wet and dry states color difference works for comparison to actual and predicted dry L*a*b* values prediction error.” For figure 7a-e. Does this baseline model based on delta-E2000 > 1.0? If yes, then add this explanation.
Answer: The baseline is not a model, it is just the raw DE2000 reading between the wet sample and the dry sample. Without a model, the raw measurement of the wet fabric is not an accurate prediction of the color when the fabric is dry. We do use DE2000 in another context as the error function in our neural-nets, so in that context we use DE2000 as an error function.
Need to provide each of the five models output/model/test parameters in a table. Regression model’s output needs a separate table.
Answer:
Information is provided in the supplemental section. See Tables S4-S7 for hyperparameters used for training the machine learning models.
Figure 3 says, pressure, temperature and speed were inputs to the neural network model, but it looks like they are not used in actual calculations.
Answer: Figure 3 is replaced to show the process described in section “Fabric Dyeing Procedure”
Page 14
Line 436: “…area is devoid of data points due to the fact these dyes are not brighteners.” – needs explanation of this part.
Answer: We have updated the text. Dyes absorb light, so there is no way to make a fabric “whiter” by dyeing. In the Lab plot, this manifests itself as a void or hole. There is also a limit to how dark we can dye a fabric, and thus all the data point in color 3-space form a sphere-like shape with a hole in the middle.
Line 436: “This distribution reflects the impact of the dyeing process, which can be influenced by factors such as fluctuations in dye concentrations, subtle variations in squeeze roller pressure, and the inherent variability of the textile fabric.” -also the dye combination is responsible.
Answer:
We have updated the sentence to include ‘dye combinations’.
“The baseline model consistently produced high CIEDE2000 error values, highlighting the inability of non-machine learning approaches to capture the complex relationships between wet and dry color states.” – so, baseline model capture only the differences between dry and wet state, and it could vary a lot due to roller pressure, given that fabric physical properties are same. So, there are valid reasons for having Delta > 1. Then why is it considered as an error? Dry state is the target state that the machine learning model predicts.
Answer:
The baseline is not a model, it is the measured color difference between the wet fabric and the dry fabric. We will try to make this clearer in the paper to avoid confusion. Because raw readings fail to accurately predict the color of the fabric in the dry state, models are needed to do so in real-time. Typically in industry, color tolerances above 0.8 to 1 are not acceptable.
Page 15: figure and table number needs correction
Line: 550: the table says it is 63.9%. not above 64%.
Answer:
The rounded value of 64% in text is now replaced with 63.9% in text.
Author Response File: Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsDr. Jasper is a well-known scholar in the field. I enjoyed reading his paper very much. The research findings from this paper could be used to improve the continuous dyeing process through real-time feedback control. I only have a few comments:
- Large amount of time was spent on dyeing samples. Please explain the reasons for not using different colored commercial samples.
- Can the model predict the color under other lighting sources?
- Please discuss the future research directions.
- Is it possible to develop a universal algorithm that can predict the dry color of fabrics regardless of fiber contents or dye class?
Author Response
Reviewer 2 Comments:
Open Review
(x) I would not like to sign my review report
( ) I would like to sign my review report
Quality of English Language
( ) The English could be improved to more clearly express the research.
(x) The English is fine and does not require any improvement.
|
Comments and Suggestions for Authors
Dr. Jasper is a well-known scholar in the field. I enjoyed reading his paper very much. The research findings from this paper could be used to improve the continuous dyeing process through real-time feedback control. I only have a few comments:
Thank you for your kind words.
- Large amount of time was spent on dyeing samples. Please explain the reasons for not using different colored commercial samples.
Answer:
We acknowledge that using commercially dyed samples of different colors could have reduced the time and resources spent on dyeing. However, in this study, we dyed the fabric in-house to control the dyeing parameters and create subtle shade variations within the same color. This was essential for training a model capable of identifying these nuanced differences.
- Can the model predict the color under other lighting sources?
Answer:
The current model can only predict/map colors under D65 lighting. To account for variations in lighting in real-time measurements under different lighting sources, reflectance spectra would need to be included in the training data. Incorporating such data would enable the model to generalize better across various lighting conditions, although it is unclear exactly how much extra data would be needed to achieve good results. This is a good point for future study.
Different light sources could result in metamerism with certain dyes which the model does not account for.
- Please discuss the future research directions.
Answer: The Limitations section is now updated to “Limitations and future directions” to include the text below:
Future research will focus on improving the model’s generalizability. This will require systematic data collection encompassing various dyeing machines, process parameters, ambient conditions, material properties, and end-use requirements. Additionally, developing a modular neural network architecture—where specialized models handle specific tasks—may enhance both generalizability and real-time accuracy, enabling the model to operate effectively at production speed. Also, it might be worth looking at the effects of metamerism and different lighting conditions.
- Is it possible to develop a universal algorithm that can predict the dry color of fabrics regardless of fiber contents or dye class?
Answer:
Yes, developing a universal algorithm to predict dry fabric color across different fiber contents and dye classes is possible, provided sufficient data is available. Initially, building such a neural network would require millions of data points to capture the complex relationships between fiber type, dye class, and drying behavior. However, once trained, the learned weights from this model could be transferred to smaller, more specialized models using transfer learning. This would allow the model to be fine-tuned on a smaller dataset specific to a particular dyeing machine, optimizing it for real-world use cases.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe article, titled "A Controlled Study on Machine Learning Applications to Predict Dry Fabric Color from Wet Samples: Influences of Dye Concentration and Squeeze Pressure," aims to develop and evaluate machine learning models for predicting the dry color of fabrics based on their wet-state appearance, considering the effects of dye concentration and squeeze pressure. This topic is highly relevant to readers of Fibers journal. Therefore, publication is recommended, provided that the following revisions are made to enhance clarity and methodological detail:
1. Keep the abstract concise, focusing directly on the study’s objective and key findings.
2. Clearly outline the main experimental parameters in the dyeing process within the methodology section. Include supporting references.
3. Provide a more detailed explanation of data normalization, specifically how it influenced model performance.
4. The equations are misordered, please review and correct them.
5. Justify why the neural network outperformed the other methods.
6. Improve the graphs by refining legends and providing clearer explanations of the axis scales.
7. Strengthen the conclusion by emphasizing the practical implications of the research for the textile industry.
8. Include a brief discussion on the study’s limitations and potential directions for future research.
9. Add recent citations on machine learning applications in textile processes. Consider including a table summarizing the latest relevant publications.
10. Provide a list of symbols to enhance readability.
11. Justify the selection of the textile parameters analyzed in this study.
Author Response
Reviewer 3 Comments
Open Review
(x) I would not like to sign my review report
( ) I would like to sign my review report
Quality of English Language
( ) The English could be improved to more clearly express the research.
(x) The English is fine and does not require any improvement.
|
|
Yes |
Can be improved |
Must be improved |
Not applicable |
Does the introduction provide sufficient background and include all relevant references? |
(x) |
( ) |
( ) |
( ) |
Is the research design appropriate? |
(x) |
( ) |
( ) |
( ) |
Are the methods adequately described? |
(x) |
( ) |
( ) |
( ) |
Are the results clearly presented? |
(x) |
( ) |
( ) |
( ) |
Are the conclusions supported by the results? |
(x) |
( ) |
( ) |
( ) |
Comments and Suggestions for Authors
The article, titled "A Controlled Study on Machine Learning Applications to Predict Dry Fabric Color from Wet Samples: Influences of Dye Concentration and Squeeze Pressure," aims to develop and evaluate machine learning models for predicting the dry color of fabrics based on their wet-state appearance, considering the effects of dye concentration and squeeze pressure. This topic is highly relevant to readers of Fibers journal. Therefore, publication is recommended, provided that the following revisions are made to enhance clarity and methodological detail:
Thank you for your feedback.
- Keep the abstract concise, focusing directly on the study’s objective and key findings.
Answer:
The final two sentences of the abstract have been revised for conciseness while ensuring that the core findings are clearly communicated. The remaining details have been retained to provide essential context for the study.
- Clearly outline the main experimental parameters in the dyeing process within the methodology section. Include supporting references.
Answer:
The main experimental parameters in the dyeing process are detailed in Section 2.2 'Fabric Dyeing Procedure.'
- Provide a more detailed explanation of data normalization, specifically how it influenced model performance.
Answer:
The following text has been added to the manuscript:
“The data was normalized to ensure that all parameters (L, a*, and b* for wet fabric) were scaled within the same range of 0 to 1. This prevents features with larger numerical ranges from dominating the training process, allowing each parameter to contribute equally.
Additionally, normalization improved model convergence speed and overall performance.
Normalization was specifically applied to neural networks, as these models are highly sensitive to feature scaling. In contrast, tree-based models (e.g., random forest, XGBoost) were trained on unnormalized data, as they rely on decision rules rather than distance-based calculations, making them inherently robust to differences in feature magnitudes.”
- The equations are misordered, please review and correct them.
Answer:
The equation numbers are now updated.
- Justify why the neural network outperformed the other methods.
Answer:
The performance of a model is determined by how closely its predictions match the actual values. Neural networks, with their deep architectures and extensive trainable parameters (over 23,000 in this study), offer greater flexibility in capturing complex relationships between wet and dry color states. Unlike tree-based models, which rely on discrete partitioning of data, neural networks can learn smooth, non-linear transformations, making them particularly effective for color prediction tasks. Additionally, normalization of input data facilitated better convergence, and if CIEDE2000 was used as a loss function, it further aligned the training process with human perceptual differences, contributing to superior performance.
- Improve the graphs by refining legends and providing clearer explanations of the axis scales.
Answer:
The figure legends are now updated:
Figure 4: Color gamut for 762 samples in the CIE 1931 color space (CIE xyY) chromaticity plot. The x-axis represents the chromaticity coordinate x (ranging from 0 to 0.8), and the y-axis represents the chromaticity coordinate y (ranging from 0 to 0.9), showing the distribution of colors in the dataset.
Figure 5: Figure 5. 3D plots of 762 color samples in the wet (a) and dry (b) states within the CIELAB (Lab) color space. The plots illustrate the shift in color coordinates due to the drying process, where L* represents lightness, while a* and b* indicate chromaticity along the red-green and blue-yellow axes, respectively.
Figure 9. Rotated views of predicted wet L*, a*, b* samples across all pressure levels. 'Blue' points represent samples with ΔE2000 (CIEDE2000 color difference) values ≤ 1, indicating high prediction accuracy. Red points represent samples with ΔE2000 values > 1, indicating greater color deviation. The L* axis represents lightness, while the a* and b* axes indicate chromaticity along the red-green and blue-yellow directions, respectively.
- Strengthen the conclusion by emphasizing the practical implications of the research for the textile industry.
Answer: Thank you for your suggestion. The last sentence of the Conclusions section is updated to emphasize the practical implications of this research in enabling real-time feedback control in continuous dyeing.
- Include a brief discussion on the study’s limitations and potential directions for future research.
Answer: Please see section 4.1 on limitations. We added a paragraph on future work.
- Add recent citations on machine learning applications in textile processes. Consider including a table summarizing the latest relevant publications.
Answer:
Add a sentence to Introduction: We have published a comprehensive review:
Ingle N, Jasper WJ. A review of deep learning and artificial intelligence in dyeing, printing and finishing. Textile Research Journal. 2024;95(5-6):625-657. doi:10.1177/00405175241268619
- Provide a list of symbols to enhance readability.
Answer:
We already have it in the section “Abbreviations”.
- Justify the selection of the textile parameters analyzed in this study.
Answer: The selection of textile parameters analyzed in this study are ones that are common throughout the textile industry. These include the choice of cotton fabric, the dyes, and the dyeing procedures.