Next steps to improve SPADE + AdaIN architecture

[51N84D]

trafficstars

We've converged to a SPADE + MUNIT hybrid, but we need some tricks to improve. Here are the short term ideas to try:

• [x] Gram Matrix as measure of style
• [x] Feature matching loss on global discriminator
• [ ] Give style information directly to spade (via addition, for example)
• [x] Use decaying non-binary masks for smoothing
• [ ] Use image of only water for perceptual loss
• [x] Compute perceptual loss on intersection of ground + flood mask

[fwextensions]

Some resources:

• https://circleci.com/blog/monorepo-dev-practices/
• https://medium.com/@mattklein123/monorepos-please-dont-e9a279be011b

[zoobot]

What side do you fall on? pro or con mono-repo?

[fwextensions]

At first glance, having all the code from one org, especially one as enormous as Google, in one repo seems crazy, but I don't have enough experience to have a strong opinion. But for a smaller project like WTT, it may make sense, especially when working on both the back and front ends, like changing the id param. Having everything in one repo would allow an atomic commit to change everything at once.