Domain Randomization for Neural Network Classification
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
Due to technical limitations, full-text HTML conversion of this manuscript could not be completed. However, the manuscript can be downloaded and accessed as a PDF.
Posted 22 Sep, 2020
On 26 Oct, 2020
Received 26 Oct, 2020
Received 09 Oct, 2020
On 01 Oct, 2020
On 28 Sep, 2020
On 28 Sep, 2020
Invitations sent on 27 Sep, 2020
On 18 Sep, 2020
On 17 Sep, 2020
On 17 Sep, 2020
On 14 Sep, 2020
Domain Randomization for Neural Network Classification
Posted 22 Sep, 2020
On 26 Oct, 2020
Received 26 Oct, 2020
Received 09 Oct, 2020
On 01 Oct, 2020
On 28 Sep, 2020
On 28 Sep, 2020
Invitations sent on 27 Sep, 2020
On 18 Sep, 2020
On 17 Sep, 2020
On 17 Sep, 2020
On 14 Sep, 2020
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
Due to technical limitations, full-text HTML conversion of this manuscript could not be completed. However, the manuscript can be downloaded and accessed as a PDF.