The new inpainting script left some ringing artefacts, I think they might come from np.where() results ordering (and the resulting imposed order of inpainting). I've tried numpy.random.shuffle(). It's still running, so I have yet to see whether this worked.
Updated -- no, it didn't. The weird lines and gradients are the edge effects of rectangular mask shapes, which I didn't like from the beginning. On the other hand, given that the inpainted values vary by ~4 counts (the contrast is stretched in this image), I'm not going to worry about this.
No comments:
Post a Comment