Description
Question 1 [25 Points] (Image Inpainting by Masked CNN) In this question, we implement a basic
masked CNN (similar to PixelCNN) for image inpainting. The image inpainting refers to the generation
of missed pixels in an image, which is considered as one of the key applications of visual generative
model; see [3] for instance.
1. Complete Question 1 in Lastname_Firstname_Asgn2.ipynb to implement a masked CNN model
for autoregressive generation.
2. Answer all parts marked by # COMPLETE .
Question 2 [25 Points] (Real NVP Implementation) In this question, we implement a basic form of
Real NVP with fully-connected scaling and translation models for MNIST image generation. You can
learn more on Real NVP in [1], though what we had in the lecture is enough to complete this question.
1. Complete Question 2 in Lastname_Firstname_Asgn2.ipynb .
2. Answer all parts marked by # COMPLETE .
References
[1] Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real nvp. In
International Conference on Learning Representations (ICLR), 2017.
[2] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter.
GANs trained by a two time-scale update rule converge to a local Nash equilibrium. In Advances in
Neural Information Processing Systems (NeurIPS), volume 30, 2017. Available at https://arxiv.org/
abs/1706.08500.
[3] Guilin Liu, Fitsum A Reda, Kevin J Shih, Ting-Chun Wang, Andrew Tao, and Bryan Catanzaro. Image
inpainting for irregular holes using partial convolutions. In Proceedings of the European Conference on
Computer Vision (ECCV), pages 85–100, 2018.