Francois Lanusse

My bad, it should be fine now.

Thanks @ismael-mendoza and @thuiop for flagging this! I'm on it

Ok, I found the problem, the free git-lfs storage of the `McWilliamsCenter` organization has ran out of bandwidth for the month (I'll notify @rmandelb here as well as this kind...

Code in branch `pixelcnn` seems to be working at least for single band observations and without specifying PSFs: 

So what we did with Fred is to extract a centered postage stamp from the frame, and that's what's fed to the pixelcnn. The position of the postage stamp is...

This being said, we thought of a few other problems. - Because the training set postage stamps are of even size, there might be a one pixel offset between the...

Hum... I don't think normalizing peak intensity is practical because it would change the level of the background noise in the images. But, if we don't particularly constrain the normalization...

But Fred's point is that the Lipschitz constant of the prior term is 10^4 larger (or smaller, I don't remember which way it goes) thank that of the likelihood term....

I have relocated development to a new branch `u/EiffL/pixelcnn` where I'm changing a little bit the prototype we had built last week with Fred. And it's now based on the...

I have packaged one of the trained pixelcnns I've been using and hosted it on the scarlter-pixelcnn repo. Here is how one would use this stuff: ```python import scarlet from...