See the published version of this preprint at Journal of Big Data.
This is a preprint, a preliminary version of a manuscript that has not completed peer review at a journal. Research Square does not conduct peer review prior to posting preprints. The posting of a preprint on this server should not be interpreted as an endorsement of its validity or suitability for dissemination as established information or for guiding clinical practice.
Learn more about In Review
Research
Svetozar Zarko Valtchev
York University
Corresponding Author
ORCiD: https://orcid.org/0000-0002-6149-138X
License:
This work is licensed under a CC BY 4.0 License. Read Full License
Large data requirements are often the main hurdle in training neural networks. Synthetic data is a cheap and efficient solution to assemble such large datasets. Using domain randomization, we show that a sufficiently well generated synthetic image dataset can be used to train a neural network classifier, achieving accuracy levels as high as 88% on 2 category classification. We show that the most important domain randomization parameter is a large variety of subjects, while secondary parameters such as lighting and textures are not. Based on our results, there is reason to believe that models trained on domain randomized images transfer to new domains better than those trained on real photos. Model performance seems to diminish slightly as the number of categories increases.
Keywords
domain randomization, synthetic image generation, neural network classifiers
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
This preprint is available for download as a PDF.
Loading...
Badges
Prescreen
This manuscript passed our Prescreen assessment
Complete author information
Appropriate declaration statements
Potential risks to human health
Prescreen
Peer Review Timeline
CURRENT STATUS: Published
View the published article at Journal of Big Data.
Version 1
Posted 22 Sep, 2020
No community comments so far
Editorial decision: Major Revision
On 12 Feb, 2021
Review #3 received
Received 10 Feb, 2021
Review #2 received
Received 10 Feb, 2021
Reviewer #4 agreed
On 26 Oct, 2020
Review #4 received
Received 26 Oct, 2020
Review #1 received
Received 09 Oct, 2020
Reviewer #3 agreed
On 01 Oct, 2020
Reviewer #1 agreed
On 28 Sep, 2020
Reviewer #2 agreed
On 28 Sep, 2020
Reviewers invited
Invitations sent on 27 Sep, 2020
Editor assigned
On 18 Sep, 2020
Submission checks complete
On 17 Sep, 2020
Editor invited
On 17 Sep, 2020
First submitted
On 14 Sep, 2020
Metrics
Comments: 0
PDF Downloads: ...
HTML Views: ...
Learn more about our company.
See the published version of this preprint at Journal of Big Data.
This is a preprint, a preliminary version of a manuscript that has not completed peer review at a journal. Research Square does not conduct peer review prior to posting preprints. The posting of a preprint on this server should not be interpreted as an endorsement of its validity or suitability for dissemination as established information or for guiding clinical practice.
Learn more about In Review
Research
Svetozar Zarko Valtchev
York University
Corresponding Author
ORCiD: https://orcid.org/0000-0002-6149-138X
Badges
Prescreen
This manuscript passed our Prescreen assessment
Complete author information
Appropriate declaration statements
Potential risks to human health
Prescreen
Peer Review Timeline
CURRENT STATUS: Published
View the published article at Journal of Big Data.
Version 1
Posted 22 Sep, 2020
No community comments so far
Editorial decision: Major Revision
On 12 Feb, 2021
Review #3 received
Received 10 Feb, 2021
Review #2 received
Received 10 Feb, 2021
Reviewer #4 agreed
On 26 Oct, 2020
Review #4 received
Received 26 Oct, 2020
Review #1 received
Received 09 Oct, 2020
Reviewer #3 agreed
On 01 Oct, 2020
Reviewer #1 agreed
On 28 Sep, 2020
Reviewer #2 agreed
On 28 Sep, 2020
Reviewers invited
Invitations sent on 27 Sep, 2020
Editor assigned
On 18 Sep, 2020
Submission checks complete
On 17 Sep, 2020
Editor invited
On 17 Sep, 2020
First submitted
On 14 Sep, 2020
Metrics
Comments: 0
PDF Downloads: ...
HTML Views: ...
License:
This work is licensed under a CC BY 4.0 License. Read Full License
View the published article at Journal of Big Data.
Large data requirements are often the main hurdle in training neural networks. Synthetic data is a cheap and efficient solution to assemble such large datasets. Using domain randomization, we show that a sufficiently well generated synthetic image dataset can be used to train a neural network classifier, achieving accuracy levels as high as 88% on 2 category classification. We show that the most important domain randomization parameter is a large variety of subjects, while secondary parameters such as lighting and textures are not. Based on our results, there is reason to believe that models trained on domain randomized images transfer to new domains better than those trained on real photos. Model performance seems to diminish slightly as the number of categories increases.
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
This preprint is available for download as a PDF.
Loading...