# 消除 Artifact，用缩放卷积神经网络生成高清图像（TensorFlow 代码）

【新智元导读】非常仔细地观察神经网络生成的图像时，经常会看到一些奇怪的棋盘格子状的伪影（棋盘效应，checkboard artifacts）。本文作者讨论了棋盘效应出现的原因以及反卷积难以避免棋盘效应的原因，最后作者提出了一种避免棋盘效应的替代方法（缩放卷积），并提供了在TensorFlow上实现的代码。

【译注：反卷积有许多解释和不同的术语，包括“转置卷积”（transposed convolution）；为了简洁起见，本文使用“反卷积”（deconvolution），有关反卷积的更多讨论参见：Dumoulin & Visin, 2016 and Shi, et al., 2016a.】

（注：在原文网站上可以调节图示的大小（size）和步长（stride）更直观地理解，下同）

#### 代码：

【译注：原文标题是Deconvolution and Checkerboard Artifacts，本文中”Checkerboard Artifact “翻译成“棋盘效应”，“Artifact”视情况翻译成“棋盘效应”或“伪影”。】

• Dong, C., Loy, C.C., He, K. and Tang, X., 2014. Image super-resolution using deep convolutional networks.arXiv preprint arXiv:1501.00092.
• Dumoulin, V., Belghazi, I., Poole, B., Lamb, A., Arjovsky, M., Mastropietro, O. and Courville, A., 2016.Adversarially Learned Inference. arXiv preprint arXiv:1606.00704.
• Dumoulin, V. and Visin, F., 2016. A guide to convolution arithmetic for deep learning. arXiv preprint arXiv:1603.07285.
• Donahue, J., Krähenbühl, P. and Darrell, T., 2016. Adversarial Feature Learning. arXiv preprint arXiv:1605.09782.
• Gauthier, J., 2015. Conditional generative adversarial networks for convolutional face generation. Technical report.
• Hénaff, O. J., & Simoncelli, E. P. (2015). Geodesics of learned representations. arXiv preprint arXiv:1511.06394.
• Johnson, J., Alahi, A. and Fei-Fei, L., 2016. Perceptual losses for real-time style transfer and super-resolution. arXiv preprint arXiv:1603.08155.
• Mordvintsev, A., Olah, C., & Tyka, M. (2015). Inceptionism: Going deeper into neural networks. Google Research Blog.
• Mordvintsev, A., 2016. DeepDreaming with TensorFlow. Github.
• Radford, A., Metz, L. and Chintala, S., 2015. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434.
• Salimans, T., Goodfellow, I., Zaremba, W., Cheung, V., Radford, A. and Chen, X., 2016. Improved techniques for training GANs. arXiv preprint arXiv:1606.03498.
• Shi, W., Caballero, J., Theis, L., Huszar, F., Aitken, A., Ledig, C. and Wang, Z., 2016. Is the deconvolution layer the same as a convolutional layer?. arXiv preprint arXiv:1609.07009.
• Shi, W., Caballero, J., Huszár, F., Totz, J., Aitken, A.P., Bishop, R., Rueckert, D. and Wang, Z., 2016. Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 1874-1883).