High Densely Connected Convolutional Networks for Denoising Monte Carlo Rendering

  • Authors

    • Mincheol Kim
    • Kwangyeob Lee
    https://doi.org/10.14419/ijet.v7i3.24.22837
  • Convolutional Neural Network, Ray Tracing, Monte Carlo Renderings, Monte Carlo Denoising, Densely Convolutional Network
  • Abstract

    Background/Objectives: Monte Carlo renderings, which are recently used in animation and visual effects, produce realistic images but noise occurs during the ray tracing process.

    Methods: In this paper, the learning is performed with only RGB channel without an auxiliary buffer such as normal, albedo, and diffuse. The performance was improved by modifying the Densely Connected Convolutional Networks, which shows excellent performance. The transition layer which has the pooling layer is removed, and the last convolution layer is used to produce a denoised image because the final layer is intended for denoising rather than classification.

    Findings: It is difficult to distinguish the detail from the noise without special information during the denoising process, thus the learning convergence speed is slowed down. However, in this paper, we found that it is possible to preserve detail while removing noise by using the Densely Connected Convolutional Network to preserve the high and low features. Even if the feature map is increased, batch normalization and bottleneck layers can resolves this problem and even increases the speed of learning. As a result, our method denoised better than state-of-the-art base-filter denoiser with only the RGB channel.

    Improvements: It was implemented by Tensor flow with Python on CPU i5 6600, and GTX 1080 Ti. After approximately 24 hours, it showed similar performance than the filter-based algorithm.

     

     

     

  • References

    1. [1] Terzopoulos, D., Platt, J., Barr, A., & Fleischer, K. (1987). Elastically deformable models. ACM SiggraphComputer Graphics, 21(4), 205-214.

      [2] Sen, P., &Darabi, S. (2012). On filtering the noise from the random parameters in Monte Carlo rendering. ACM Trans. Graph., 31(3), 18-1.

      [3] Li, T. M., Wu, Y. T., & Chuang, Y. Y. (2012). SURE-based optimization for adaptive sampling and reconstruction. ACM Transactions on Graphics (TOG), 31(6), 194.

      [4] Moon, B., Carr, N., & Yoon, S. E. (2014). Adaptive rendering based on weighted local regression. ACM Transactions on Graphics (TOG), 33(5), 170.

      [5] Kalantari, N. K., Bako, S., & Sen, P. (2015). A machine learning approach for filtering Monte Carlo noise. ACM Trans. Graph., 34(4), 122-1.

      [6] Bako, S., Vogels, T., McWilliams, B., Meyer, M., Novák, J., Harvill, A., ...&Rousselle, F. (2017). Kernel-predicting convolutional networks for denoising Monte Carlo renderings. ACM Transactions on Graphics (TOG), 36(4), 97.

      [7] Witten, I. H., Frank, E., Hall, M. A., & Pal, C. J. (2016). Data Mining: Practical machine learning tools and techniques. Morgan Kaufmann.

      [8] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778)

      [9] Huang, G., Liu, Z., Weinberger, K. Q., & van der Maaten, L. (2017, July). Densely connected convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (Vol. 1, No. 2, p. 3).

      [10] Zhao, H., Gallo, O., Frosio, I., &Kautz, J. (2017). Loss functions for image restoration with neural networks. IEEE Transactions on Computational Imaging, 3(1), 47-57.

      [11] He, K., Zhang, X., Ren, S., & Sun, J. (2015). Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. In Proceedings of the IEEE international conference on computer vision (pp. 1026-1034).

  • Downloads

  • How to Cite

    Kim, M., & Lee, K. (2018). High Densely Connected Convolutional Networks for Denoising Monte Carlo Rendering. International Journal of Engineering & Technology, 7(3.24), 662-665. https://doi.org/10.14419/ijet.v7i3.24.22837

    Received date: 2018-12-02

    Accepted date: 2018-12-02