Generative Adversarial Text to Image Synthesis

Because that isn't true? Looking at the first thing (GANs): developed by Ian Goodfellow, who is now at OpenAI. The first linked paper (https://arxiv.org/pdf/1605.05396v2.pdf) is out of University of Michigan and Max Planck Institute. The second (https://phillipi.github.io/pix2pix/) is out of Berkeley, the third (https://arxiv.org/pdf/1609.04802v3.pdf) is out of Twitter (which I guess counts as a rich American tech company). Google Brain/DeepMind/FAIR/MS Research do great work, but there is plenty of great work elsewhere too. For example, just yesterday an implementation[1] of Fast Layer Normalization for TensorFlow was posted on Reddit[2] but someone, who says they don't really do C++ ("I am really new to CUDA and C++"). This can speed up training (sometimes) by 5-10X (!). That's democratization. [1] https://github.com/MycChiu/fast-LayerNorm-TF [2] https://www.reddit.com/r/MachineLearning/comments/5gt0wm/p_f...

This article is best juxtaposed with articles about image synthesis: * Wave Function Collapse image generator https://github.com/mxgmn/WaveFunctionCollapse * Image Synthesis from Yahoo's open_nsfw: https://open_nsfw.gitlab.io * Generative Adversarial Text to Image Synthesis: https://arxiv.org/abs/1605.05396 Put another way, when I look at the McMansions at 10 on the scale, it looks to me like the kind of output you get when you just feed a bunch of pictures of houses into a neural network without feeding in any of the cultural or aesthetic context. Things like the large garages and roofs look like the efforts of constraint solvers working with inputs that are out of bounds. In a sense, that is exactly what happened.

> One main thing it lacks is imagination. [...] For example, if I ask you to imagine a Guitar playing Dolphin, you could imagine it and even recognize it from a cartoon even though you have never seen it in your life before. Not so for Deep Learning These days one should be careful when claiming things deep learning can't do. There are in fact systems that can imagine things it has never seen before. Here's an example. It takes any text, and imagines an image for it: https://github.com/emansim/text2image Here's a more recent example, with even higher quality images: http://arxiv.org/pdf/1605.05396v2.pdf