Synthetic Dataset Generation Using Photo-Realistic Simulation with Varied Time and Weather Axes

Round 1

Reviewer 1 Report

Abstract is confusing, authors must revise and rewrite the abstract for conciseness and clarity.

Language must be revised by a native English speaker. For instance:

- “which worsens the further the input data deviates from the training data.”

In abstract, authors state:

- “… and identifies both issues …”

What are these issues?

Introduction is quite difficult to read and follow, it must be rewritten to improve its comprehension.

The organization, structure and flow of the introduction must be improved significantly to make it readable.

What is M in the following sentence?

- “if a dataset of colour images contains N samples and M,”

Please use standard symbology

- “M <= Mmax”

Figure 1 is never referenced. What is the purpose of this figure? It must be removed.

All acronyms must be defined at their first appearance in the main text. For instance:

- Digital Twin (DT).

- European Union Aviation Safety Agency (EASA).

- Unmanned Air Systems (UAS).

In the introduction, authors must clearly describe the novelty and contribution of their proposed approach, regarding previous works in the state of the art about the subject, to support their claim for a possible contribution.

The manuscript must be proofread to eliminate writing issues. For instance:

- “… using a trained trained network reading …”

There are too many writing errors in the manuscript that make really difficult to read and understand its intended purpose. For instance:

- “As previously stated, Neural Networks have accuracy issues regarding operation in environments which the autonomous navigator is not trained for [21], which may not hinder autonomous UAV implementations mechanically but could in terms of of legislation.”

Section 1 Introduction and Section 2 Background must be completely rewritten and shortened looking for conciseness and completeness.

The manuscript lacks from consistecy. For instance, Figure 4 and Fig. 5

Why is figure 9 referenced in the main text before figure 7 and figure 8?

Abstract is confusing, authors must revise and rewrite the abstract for conciseness and clarity.

Language must be revised by a native English speaker

Introduction is quite difficult to read and follow, it must be rewritten to improve its comprehension.

The organization, structure and flow of the introduction must be improved significantly to make it readable.

Author Response

Thank you so much for your comments, I have addressed each comment below, for convenience I have also included an excerpt of any major changes made to the manuscript in response to the comments.

Abstract is confusing, authors must revise and rewrite the abstract for conciseness and clarity.

I have rewritten the abstract to be more clear in terms of the rationale and findings of the paper. I hope these improvements are acceptable. Thank you.

Language must be revised by a native English speaker. For instance:

- “which worsens the further the input data deviates from the training data.”

Could you please inform me as to what part of this sentence is incorrect? I am a native English speaker, and neither myself nor my colleagues could find any issue with this statement.

In abstract, authors state:

- “… and identifies both issues …”

What are these issues?

This comment in the abstract refers to the issues in the simulation detailed in the results section, however after rewriting I have deemed this to be trivial information which is not relevant for the abstract. As such it has been removed.

What is M in the following sentence?

- “if a dataset of colour images contains N samples and M,”

Please use standard symbology

- “M <= Mmax”

M is the theoretical maximum amount of unique information contained within a dataset. I have rewritten this section to more clearly explain this variable and to fix the formula regarding Mmax, thank you:

"...if a dataset of colour images contains N samples and M, a theoretical amount of unique information, then the average unique information per sample L = M/N. Theoretically, if augmentation is used to inflate the dataset through the injection of noise, image skewing, or other common transformative methods, the number of samples in the dataset increases but the amount of unique information in the dataset remains the same, unless new information is injected during the augmentation process. This results in more samples, but lower unique information per sample, which is a problem that is solvable through simulation. By using simulation, a researcher is capable of creating new samples while maintaining the amount of unique information per sample up to the theoretical limit L = M/N; 0 < M <= Mmax, where Mmax is the maximum unique information available within the simulation."

Now reads:

"...if a dataset of colour images contains N samples and a theoretical amount of unique information contained within the dataset, defined as M. The average unique information per sample L = M/N. Theoretically, if augmentation is used to inflate the dataset through the injection of noise, image skewing, or other common transformative methods, the number of samples in the dataset increases but the amount of unique information in the dataset remains the same, unless new information is injected during the augmentation process. This results in more samples, but lower unique information per sample, which is a problem that is solvable through simulation. By using simulation, a researcher is capable of creating new samples while maintaining the amount of unique information per sample up to a theoretical limit L = M/N; 0 < M ≤ Mmax, where Mmax is the maximum unique information contained within the simulation. Using this project as an example, Mmax can be considered to be the information from the 3D scan data and textures, all the possible environmental contexts the simulation can create, and all the possible object and camera positions and configurations, with no repeating information."

Figure 1 is never referenced. What is the purpose of this figure? It must be removed.

I have adjusted section 1 to include a reference to the first figure in the relevant location. Thank you for spotting this.

In Section 1, the line:

"By using a combination of modern 3d scanning techniques to reconstruct a known area as a Digital Twin inside of the Unity game engine..."

now reads:

"The simulator shown in Figure 1 demonstrates how the Unity game engine can be used to rapidly create a simulator that uses a combination of modern 3d scanning techniques to reconstruct an area."

All acronyms must be defined at their first appearance in the main text. For instance:

- Digital Twin (DT).

- European Union Aviation Safety Agency (EASA).

- Unmanned Air Systems (UAS).

This has been resolved, I have also gone over the manuscript again to ensure that acronyms are used more appropriately, namely removing EASA and UAS as acronyms since they only appear once.

In the introduction, authors must clearly describe the novelty and contribution of their proposed approach, regarding previous works in the state of the art about the subject, to support their claim for a possible contribution.

Thank you for this comment, I have adjusted Section 1 to include more information and reference to previous literary review on the topic, the following statement:

"Among previously identified autonomous features, Collision Avoidance, Obstacle Detection, and Object Distinction were the most popular. Only approx. 7.7\% of the projects examined over the previous 5 years consider operation in more than one environment (e.g. Navigation in an Urban environment such as a city and a Natural environment such as a forest) [5]."

Now reads:

"Previous review of state of the art solutions in the autonomous navigation research space revealed that, of the classified autonomous features, Collision Avoidance, Obstacle Detection, and Object Distinction (also known as Object Detection) were the most popular. Only approx. 7.7% of the projects examined over the 5 years prior to review considered operation in more than one environment (e.g. Navigation in an Urban environment such as a city and a Natural environment such as a forest) [5]."

The manuscript must be proofread to eliminate writing issues. For instance:

- “… using a trained trained network reading …”

I have addressed this, thank you.

There are too many writing errors in the manuscript that make really difficult to read and understand its intended purpose. For instance:

- “As previously stated, Neural Networks have accuracy issues regarding operation in environments which the autonomous navigator is not trained for [21], which may not hinder autonomous UAV implementations mechanically but could in terms of of legislation.”

Can you please explain what specifically is wrong with this sentence? Neither myself nor my colleagues could find any issue with this statement.

The manuscript lacks from consistecy. For instance, Figure 4 and Fig. 5

This has been resolved, I have checked over each figure reference to ensure that they are referred in the same way. Thank you.

Why is figure 9 referenced in the main text before figure 7 and figure 8?

This issue has been resolved, thank you.

Introduction is quite difficult to read and follow, it must be rewritten to improve its comprehension.

The organization, structure and flow of the introduction must be improved significantly to make it readable.

Section 1 Introduction and Section 2 Background must be completely rewritten and shortened looking for conciseness and completeness.

Thank you for these comments, I have adjusted the introduction in response to other reviewers comments, and I believe it to be more concise and readable now. I hope the changes are acceptable to address this concern.

Reviewer 2 Report

This paper proposes an artificial image dataset generation method using 3D simulation. In the reviewer’s opinion, the idea of this paper lacks innovation and the content of this paper lacks enough technique details. In addition, there exists many English writing errors in this paper. The comment are as follows.

1. In this paper, the influence on images of different time and weather is light level, i.e., brightness of images. Traditional image augmentation can accomplish the expected results by adding different filters. Simulations with different light level in Unity are unnecessary.

2. The purpose of the proposed method is for autonomous navigation of UAVs. However, technique details of autonomous navigation are not included in the proposed method and experiment. The connection between UAV autonomous navigation and the proposed method is ambiguous.

3. This paper claims to adopt a digital twin approach through maintaining a link to the physical space. However, what is the physical link and how the link is implemented are not included in this paper.

4. Literature review on image generation is missing.

5. This paper uses 3D scan to collect the original data, which is expensive and is not suitable for training network of autonomous navigation.

6. Object detection has been studied a lot in the field of computer vision. What’s the advantage of the proposed method?

7. The descriptions on experiment setting are missing. The simple experiment on bike detection is not convincing.

There are a few grammatical or vocabulary errors, sunce as: line 6 (worsens the further the input data deviates from the training data), line 195 (To summarize;) the legend of Fig.5 in page 8, line 333 to line 337.

Author Response

This paper proposes an artificial image dataset generation method using 3D simulation. In the reviewer’s opinion, the idea of this paper lacks innovation and the content of this paper lacks enough technique details. In addition, there exists many English writing errors in this paper. The comment are as follows.

Thank you so much for your comments, I have addressed each comment below, for convenience I have also included an excerpt of any changes made to the manuscript in response to the comments.

1. In this paper, the influence on images of different time and weather is light level, i.e., brightness of images. Traditional image augmentation can accomplish the expected results by adding different filters. Simulations with different light level in Unity are unnecessary.

I have expanded the paragraph in section 1 that discusses this, which will hopefully make this point clearer to the reader:

"The issue with Augmentation is that samples are created by recycling old information, if a dataset of colour images contains N samples and M, a theoretical amount of unique information, then the average unique information per sample L = M/N. Theoretically, if augmentation is used to inflate the dataset through the injection of noise, image skewing, or other common transformative methods, the number of samples in the dataset increases but the amount of unique information in the dataset remains the same, unless new information is injected during the augmentation process. This results in more samples, but lower unique information per sample, which is a problem that is solvable through simulation. By using simulation, a researcher is capable of creating new samples while maintaining the amount of unique information per sample up to the theoretical limit L = M/N; 0 < M <= Mmax, where Mmax is the maximum unique information available within the simulation" now reads:

"The issue with augmentation methods is that samples are created by mostly recycling old information, if a dataset of colour images contains N samples and a theoretical amount of unique information contained within the dataset, defined as M. The average unique information per sample L = M/N. Theoretically, if augmentation is used to inflate the dataset through the injection of noise, image skewing, or other common transformative methods, the number of samples in the dataset increases but the amount of unique information in the dataset remains the same, unless new information is injected during the augmentation process. This results in more samples, but lower unique information per sample, which is a problem that is solvable through simulation. By using simulation, a researcher is capable of creating new samples while maintaining the amount of unique information per sample up to a theoretical limit L = M/N; 0 < M ≤ Mmax, where Mmax is the maximum unique information contained within the simulation. Using this project as an example, Mmax can be considered to be the information from the 3D scan data and textures, all the possible environmental contexts the simulation can create, and all the possible object and camera positions and configurations, with no repeating information."

2. The purpose of the proposed method is for autonomous navigation of UAVs. However, technique details of autonomous navigation are not included in the proposed method and experiment. The connection between UAV autonomous navigation and the proposed method is ambiguous.

6. Object detection has been studied a lot in the field of computer vision. What’s the advantage of the proposed method?

7. The descriptions on experiment setting are missing. The simple experiment on bike detection is not convincing.

The task of object detection was chosen because of it's popularity among studies and its applicability towards autonomous navigation systems. The choice of bicycles as a target object was to demonstrate the simulation concept with a COCO recognisable object of which suitable scans could be sourced online. The object detection experiment could be of any scannable object type as long as the model considers it as a type. And the DL network used could be any network that can take an image as an input. This paper is about considering a solution to the data availability problem of image training and testing discussed in Section 1 through the simulation of 3d scans in an environment, with contextual references via the task and implementation to both Digital Twins and Autonomous Navigation.

The connection is referred to in previous work and at the start of Section 1 and is also present in Section 2.1:

I have adjusted Section 1 to include more information on the topic, the following statement:

"Among previously identified autonomous features, Collision Avoidance, Obstacle Detection, and Object Distinction were the most popular. Only approx. 7.7\% of the projects examined over the previous 5 years consider operation in more than one environment (e.g. Navigation in an Urban environment such as a city and a Natural environment such as a forest) [5]."

Now reads:

"Previous review of state of the art solutions in the autonomous navigation research space revealed that, of the classified autonomous features, Collision Avoidance, Obstacle Detection, and Object Distinction (also known as Object Detection) were the most popular. Only approx. 7.7% of the projects examined over the 5 years prior to review considered operation in more than one environment (e.g. Navigation in an Urban environment such as a city and a Natural environment such as a forest) [5]."

3. This paper claims to adopt a digital twin approach through maintaining a link to the physical space. However, what is the physical link and how the link is implemented are not included in this paper.

Thank you for comment. the claim is of a theoretical digital twin link, since the asset is websourced, a digital twin link is possible with physical access to the location, but is not maintained for this experiment, I have made changes throughout the document to ensure that this is made clear wherever it is mentioned, and expanded the clarification in Section 3.1:

"It is recommended for creating an area as a DT that the scanned area is physically accessible for the purposes of verifying metrics such as dimensional accuracy, however area scanning is a highly involved process which requires special equipment and training that is outside the scope of this project, however it is intended for future research. For this proof of concept, the referred 3D Scan from Sketchfab [ 53 ] was used."

5. This paper uses 3D scan to collect the original data, which is expensive and is not suitable for training network of autonomous navigation.

Thank you for your comment. 3D scan data can be expensive, but solutions exists which are cheaper, even free (Meshroom is open source and more recently RealityCapture has produced a freeware version of its software). 3D scanning hardware is expensive, but scans can be created using video footage and photographs (photogrammetry) from a decent smartphone camera. I have included this as a point in Section 3.1.

There are a few grammatical or vocabulary errors, sunce as: line 6 (worsens the further the input data deviates fromthe training data)

Could you please inform me as to what part of this sentence is incorrect? Neither myself nor my colleagues could find any issue with this statement.

line 195 (To summarize;) the legend of Fig.5 in page 8, line 333 to line 337.

I have resolved the issues and adjusted the paragraph at line 333-337 to be more readable:

"Figure 6 is an amended flowchart designed to maximise the benefit of this property and outline of the general sampling process of a manual phase which is first used to collect the position and rotations of the camera for each sample and an automated phase which moves the camera to each position and rotation which then generates the images, The result of this process is a dataset that can be sliced in more dimensions than normal manual collection would allow."

Now reads:

"Figure 6 is an amended flowchart designed to maximise the benefit of this property and outline of the general sampling process. The process consists of a manual phase which is first used to collect the position and rotations of the camera for each sample, and an automated phase which moves the camera to each position and rotation, which then generates the images at that iteration, this process then repeats for every iteration defined in the simulation. The result of this process is a dataset that can be sliced in more dimensions than normal manual collection would allow."

4. Literature review on image generation is missing.

I have resolved this issue, thank you.

Reviewer 3 Report

In this paper, a method for generating synthetic datasets based on 3D scans of objects, followed by extensions to the dataset through variations in weather and time components, was proposed. It is a meaningful work to address issue of scaling dataset size and complexity. However, the paper presentation should be improved.

1. Always spell out the acronym the first time it is used in the abstract and body of the paper.

2. Proofread the paper to correct the typos and improve the readability, e.g., nav- igational. Don't capitalize the first character if no necessary, e.g., Autonomously. Spell out "approx." Change "3d" to "3D" in the paper including figures.

3. In the abstract, what does the "23across" mean?

4. Fig. 1 was not cited in the paper.

5. Add some bulletins to summarize the paper contribution on the end of introduction.

6. Improve the quality of Fig. 6.

7. Change the time representation from "hhmm" to "hh:mm". Otherwise, it is very confused to read the paper.

 8.  Avoid using in-text references in a conclusion.

Use a comma instead of a colon after "for example." Proofread the paper to correct typos, such as "ie" (which should be i.e.) in line 97. Additionally, the authors should consider revising the writing to avoid beginning two consecutive sentences with "however."

Author Response

In this paper, a method for generating synthetic datasets based on 3D scans of objects, followed by extensions to the dataset through variations in weather and time components, was proposed. It is a meaningful work to address issue of scaling dataset size and complexity. However, the paper presentation should be improved.

1. Always spell out the acronym the first time it is used in the abstract and body of the paper.

This issue has been raised by another reviewer and has been resolved, thank you.

3. In the abstract, what does the "23across" mean?

This was a latex typo which caused the sentence following the 23(meant to be 23%) to break, and has since been resolved as I have rewritten the abstract. Thank you.

4. Fig. 1 was not cited in the paper.

I have adjusted section 1 to include a reference to the first figure in the relevant location. Thank you for spotting this.

In Section 1, the line:

"By using a combination of modern 3d scanning techniques to reconstruct a known area as a Digital Twin inside of the Unity game engine..."

now reads:

"The simulator shown in Figure 1 demonstrates how the Unity game engine can be used to rapidly create a simulator that uses a combination of modern 3d scanning techniques to reconstruct an area."

5. Add some bulletins to summarize the paper contribution on the end of introduction.

Thank you for the suggestion, I have added three points regarding the contribution to the end of Section 1:

"To summarise, the contribution of this paper is as follows:

• Outlines a novel method for the generation of synthetic samples for image based training by using 3D-Scanning and a photorealistic game engine.
• Simulates weather and light changes to create a contextually varied image dataset for the potential training and testing of image based autonomous navigators.
• Tests the environmentally varied dataset against a commonly used generic YoloV3 Object Detection model, and compares it to a realistic baseline to demonstrate the consistency in simulation performance against real image data."

6. Improve the quality of Fig. 6.

Figure 6 has been adjusted to provide more detail, and the resolution of the image has been improved, thank you.

7. Change the time representation from "hhmm" to "hh:mm". Otherwise, it is very confused to read the paper.

I have adjusted all the time references in the document to this format, thank you for pointing this out.

8. Avoid using in-text references in a conclusion.

Thank you for pointing this out, I haven't heard of this convention and have resolved the issue, I have kept the first two references in the conclusion since they are in the first two paragraphs which is simply a reiteration of the scope from previous work, I believe this restatement reads better than simply jumping in to the conclusion.

2. Proofread the paper to correct the typos and improve the readability, e.g., nav- igational. Don't capitalize the first character if no necessary, e.g., Autonomously. Spell out "approx." Change "3d" to "3D" in the paper including figures.

Use a comma instead of a colon after "for example." Proofread the paper to correct typos, such as "ie" (which should be i.e.) in line 97. Additionally, the authors should consider revising the writing to avoid beginning two consecutive sentences with "however."

Thank you for these comments, I have made adjustments throughout the manuscript to resolve the typos, and have also restructured the end paragraph of Section 4.1 to address the last issue:

"At peak conditions the model is detecting as normal, however when the harmonic mean is considered a single object was drastically harder to detect than the others. This is not a surprising result by itself, however the response for Bike 3 is normal. This is strange given that Bike 3 and Bike 4 are the same model of children’s bicycle, with Bike 3 being a green version and Bike 4 being a pink version."

Now reads:

"At peak conditions the model is detecting as normal, when the harmonic mean is considered Bike 4 was drastically harder to detect than the others. This is not a surprising result by itself, however the response for Bike 3 is normal. This is strange given that Bike 3 and Bike 4 are the same model of children’s bicycle, with green and pink frames respectively"

Reviewer 4 Report

The paper mentions using the Digital Twin method to construct simulations and comparing them with real datasets. However, there is no detailed explanation of how the digital twin method is applied to this study, and there is no reason why it is more efficient or informative compared to traditional experimental methods.

 When describing the collection process of 3D scanning resources, it was mentioned that five assets were obtained from the same repository, and tables and image comparisons of these assets were provided. However, there was no explanation as to why these specific assets were chosen, nor was more information provided about their diversity, representativeness, and applicability.

Dataset evaluation: Although the results show that the simulated dataset can better simulate the impact of real-world scenarios on network response, it is recommended to further explore different combinations of timelines and weather conditions, and increase more extreme case data samples. 

The references listed are not new enough, and recent work needs to be supplemented.

The quality of English language is fine.

Author Response

The paper mentions using the Digital Twin method to construct simulations and comparing them with real datasets. However, there is no detailed explanation of how the digital twin method is applied to this study, and there is no reason why it is more efficient or informative compared to traditional experimental methods.

Thank you for comment. the claim is of a theoretical digital twin link, since the asset is websourced, a digital twin link is possible with physical access to the location, but is not maintained for this experiment, I have made changes throughout the document to ensure that this is made clear wherever it is mentioned, and expanded the clarification in Section 3.1:

"It is recommended for creating a Digital Twin area that the scanned area is physically accessible for the purposes of verifying metrics such as dimensional accuracy, however area scanning is a highly involved process which requires special equipment and training that is outside the scope of this project, however it is intended for future research. For this proof of concept, the referred 3D Scan from Sketchfab [ 53 ] was used."

Now reads:

"While it is possible to recreate large area scans without expensive equipment through the use of commercially available or open source photogrammetry tools and decent smartphone cameras, due to time constraints it was considered outside the achievable scope of this project. The potential for maintaining a DT approach in the simulation remains possible by using spatial measurements of physically accessible locations as a DT link, and is intended for future expansions on this project. For this proof of concept, the referred 3D Scan from Sketchfab [ 53 ] was used as a placeholder."

When describing the collection process of 3D scanning resources, it was mentioned that five assets were obtained from the same repository, and tables and image comparisons of these assets were provided. However, there was no explanation as to why these specific assets were chosen, nor was more information provided about their diversity, representativeness, and applicability.

Thank you for this comment, I have expanded both the text description and Table 1 to provide additional detail on this.

"For this experiment five assets were sourced from the same repository as the scene geometry [ 54 – 57], a description of these assets can be found in Table 1 and represents an acceptable spread of different bicycle types by two defining factors, that being

the frame size and tires of the object."

Dataset evaluation: Although the results show that the simulated dataset can better simulate the impact of real-world scenarios on network response, it is recommended to further explore different combinations of timelines and weather conditions, and increase more extreme case data samples. 

Thank you for this comment, more extreme cases were considered for future research, section 4.2 has been expanded to include more information on this:

"Though initially included in prior work, 'Fog' and 'Snow' weather configurations are not included in the simulation or subsequent simulated data sets due to poor simulation performance and lack of realistic image samples for comparison, though these configurations are a potential topic for future work."

Now reads:

"Though initially included in prior work, 'Fog' and 'Snow' weather configurations are not included in the simulation or subsequent simulated data sets due to poor simulation performance and lack of realistic image samples for comparison, however more bespoke simulation attempts could consider these elements along with many other environmental factors as variables to be combined with other effects and as such is considered a viable topic for future research."

The references listed are not new enough, and recent work needs to be supplemented.

I have adjusted some of the older references with work that contains more up to date information, thank you.

Round 2

Reviewer 1 Report

Figure 1 is not referenced in the main text yet.

Many language mistakes are still present in the manuscript. The manuscript must be revised by a native English speaker.

- For instance:

“… but could in terms of of legislation …”

The manuscript is still very difficult to read.

Authors claim to have addressed the suggested comments and suggestions; however, these are not reflected in the revised version of the manuscript.

From this reviewer point of view, Section 1 and Section 2 are still too long and quite difficult to read and follow.

The manuscript is still very difficult to read.

Author Response

Figure 1 is not referenced in the main text yet.

Thank you for your comment. In the previous round I adjusted section 1 to include a reference to the first figure in the relevant location.

Section 1, Line 74, reads:

"The simulator shown in Figure 1 demonstrates how the Unity game engine can be used to rapidly create a simulator that uses a combination of modern 3d scanning techniques to reconstruct an area."

Many language mistakes are still present in the manuscript. The manuscript must be revised by a native English speaker.

- For instance:

“… but could in terms of of legislation …”

Thank you for pointing out this issue, I have resolved the typo in this sentence.

The manuscript is still very difficult to read.

Authors claim to have addressed the suggested comments and suggestions; however, these are not reflected in the revised version of the manuscript.

From this reviewer point of view, Section 1 and Section 2 are still too long and quite difficult to read and follow.

I am sorry you did not find the previous changes sufficient. I have gone over sections 1 and 2 in finer detail with the intention of reducing these sections while maintaining clarity and reinforcing the narrative. I believe I have improved the readability of these sections significantly while removing almost a full page from the manuscript. Since there are many changes to sections 1 and 2, for your convenience I have included a pdf in this response with the highlighted changes to these sections. I hope you find this acceptable. Thank you.

Author Response File: Author Response.pdf

Reviewer 4 Report

The authors have revised and improved the manuscript, and responsed the comments of the reviewer.

The quality of English language is acceptable.

Author Response

Thank you for validating that the changes in the previous round were sufficient. I have gone over the manuscript again in order to improve some of the language and readability. For your convenience I have included a pdf in this response with the highlighted changes in the new manuscript.

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

No further comments

No further comments