- Select a paper [1-12] to present in class. Must be approved by instructor.
- Deadline: 23/09
- Prepare slides (30 min.) for a class presentation (dates are on Schedule).
- Focus on: i) key ideas; ii) contribution to SoTA.
- Deadline: 21/10
====================================================================== I - Improving GANS === Quality / Resolution ---* Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks (DC-GAN) [1] https://arxiv.org/abs/1511.06434 https://github.com/Newmu/dcgan_code ---* Progressive Growing of GANs for Improved Quality, Stability, and Variation (Pro-GAN) [2] https://arxiv.org/pdf/1710.10196v3.pdf https://github.com/tkarras/progressive_growing_of_gans === Stability / Metrics ---* Wasserstein GAN (W-GAN) [3] https://arxiv.org/abs/1701.07875 https://github.com/martinarjovsky/WassersteinGAN ---* Spectral Normalization for Generative Adversarial Networks (SN-GAN) [4] https://arxiv.org/abs/1802.05957 https://github.com/godisboy/SN-GAN ====================================================================== II - Content == Conditional ---* Gan Dissection: Visualizing And Understanding Generative Adversarial Networks [5] https://openreview.net/pdf?id=Hyg_X2C5FX https://github.com/CSAILVision/GANDissect https://gandissect.csail.mit.edu ---* Large Scale GAN Training for High Fidelity Natural Image Synthesis (BigGAN) [6] https://arxiv.org/abs/1809.11096 https://github.com/ajbrock/BigGAN-PyTorch == Structure / Style --* A Style-Based Generator Architecture for Generative Adversarial Networks (StyleGAN) [7] https://arxiv.org/abs/1812.04948 https://github.com/NVlabs/stylegan --* Analyzing and Improving the Image Quality of StyleGAN (StyleGAN2) [8] https://arxiv.org/abs/1912.04958 https://github.com/NVlabs/stylegan2-ada-pytorch ====================================================================== III - Authoring == Transformation / Self-Supervision ---* Image-to-Image Translation with Conditional Adversarial Networks (pix2pix) [9] https://arxiv.org/abs/1611.07004 https://phillipi.github.io/pix2pix/ ---* Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks (CycleGAN) [10] https://arxiv.org/pdf/1703.10593.pdf https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix == Editing ---* Semantic Image Synthesis with Spatially-Adaptive Normalization (GauGAN) [11] https://arxiv.org/abs/1903.07291 https://github.com/nvlabs/spade/ ---* Anycost GANs for Interactive Image Synthesis and Editing [12] https://arxiv.org/abs/2103.03243 https://github.com/mit-han-lab/anycost-gan ====================================================================== IV - 3D ---pi-GAN: Periodic Implicit Generative Adversarial Networks for 3D-Aware Image Synthesis https://arxiv.org/abs/2012.00926 https://github.com/marcoamonteiro/pi-GAN ---Image GANs meet Differentiable Rendering for Inverse Graphics and Interpretable 3D Neural Rendering https://arxiv.org/abs/2010.09125 ----Deep View Synthesis via Self-Consistent Generative Network https://arxiv.org/abs/2101.10844 ====================================================================== Etc. ---* A Learned Representation For Artistic Style (Conditional) https://arxiv.org/abs/1610.07629 https://github.com/joelmoniz/gogh-figure ======================================================================