Sometimes we see artifacts in semantic segmentation networks. These neural networks often need a way to upsample feature maps which can be a source of such noise. There are a few papers on this noise which I will summarize here, including a simple, promising, (relatively) new approach in the last paper (with link to an implementation). Lastly, I list a few other potential areas to look when encountering this problem.
Posts by Tags
In 2018, I published a blogpost about my VIM setup. Since I’m continously tweaking my setup, my previous post is quite outdated from my current setup. This means it is time for an update.