De-noise-GAN: De-noising Images to Improve RoboCup Soccer Ball Detection
A moving robot or moving camera causes motion blur in the robot’s vision and distorts recorded images. We show that motion blur, differing lighting, and other distortions heavily affect the object localization performance of deep learning architectures for RoboCup Humanoid Soccer scenes. The paper proposes deep conditional generative models to apply visual noise filtering. Instead of generating new samples for a specific domain our model is constrained by reconstructing RoboCup soccer images. The conditional DCGAN (deep convolutional generative adversarial network) works semi-supervised. Thus there is no need for labeled training data. We show that object localization architectures significantly drop in accuracy when supplied with noisy input data and that our proposed model can significantly increase the accuracy again.
KeywordsTensorFlow Neural networks DCGAN GAN De-noising RoboCup Robotics
We are grateful to the NVIDIA corporation for supporting our research through the NVIDIA GPU Grant Program (https://developer.nvidia.com/academic_gpu_seeding). We used the donated NVIDIA Titan X (Pascal) to train our models. The work was made in collaboration with the TRR 169 “Crossmodal Learning”, funded by the DFG.
- 1.Aitken, A., Ledig, C., Theis, L., Caballero, J., Wang, Z., Shi, W.: Checkerboard artifact free sub-pixel convolution: a note on sub-pixel convolution, resize convolution and convolution resize, July 2017. http://arxiv.org/abs/1707.02937
- 2.Dahl, R., Norouzi, M., Shlens, J.: Pixel recursive super resolution. arXiv preprint arXiv:1702.00783 (2017)
- 3.Goodfellow, I., et al.: Generative adversarial nets. In: Advances in Neural Information Processing Systems, vol. 27, pp. 2672–2680 (2014). http://papers.nips.cc/paper/5423-generative-adversarial-nets.pdf
- 4.Karpathy, A., Leung, T.: Large-scale video classification with convolutional neural networks. In: Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition, pp. 1725–1732 (2014). https://doi.org/10.1109/CVPR.2014.223
- 5.Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet classification with deep convolutional neural networks. In: Advances In Neural Information Processing Systems, pp. 1–9 (2012)Google Scholar
- 6.Ledig, C., et al.: Photo-realistic single image super-resolution using a generative adversarial network, September 2016. http://arxiv.org/abs/1609.04802
- 7.Mishkin, D., Sergievskiy, N., Matas, J.: Systematic evaluation of CNN advances on the ImageNet, June 2016. http://arxiv.org/abs/1606.02228
- 9.Radford, A., Metz, L., Chintala, S.: Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, pp. 1–16 (2016)
- 10.Szegedy, C., Reed, S., Sermanet, P., Vanhoucke, V., Rabinovich, A.: Going deeper with convolutions, pp. 1–12 (2014)Google Scholar
- 11.Zhang, H., Sindagi, V., Patel, V.M.: Image de-raining using a conditional generative adversarial network. arXiv preprint arXiv:1701.05957, pp. 1–13 (2017)