Christoph Heindl

I think you should look into models that can predict missing pixels for entire images at once: See here for example https://arxiv.org/abs/1905.01639

Ah ok, I was already planning for variance reduction methods :) For larger sigmae everything seems to be much smoother - that I observed as well. I wonder if the...

Yes, very true if data dimensions become large. I was thinking about (low-rank) approximations to the jacobian and came across this paper > Abdel-Khalik, Hany S., et al. "A low...

I've recreated your toy-example to compare Langevin and annealed Langevin sampling. In particular, I've not used exact scores but trained a toy model to perform score prediction. The results are...

ah nice, thanks for letting me know. Will hopefully update soon.

@guoyangqin sorry I missed your question for more than 2 years :(. Yes, random cost matrices are probably not representative for a lot of real-world problems like you mention. More...

Yes you might be right that random instances are rather made up, but at least the dense case is very relevant to me. In [motmetrics](https://github.com/cheind/py-motmetrics) we are globally matching detections...

Thanks for the suggestion @kylemcdonald. This looks like a promising benchmark to add. Hopefully I find some time in the upcoming holidays to add. To understand this fully let me...

@kylemcdonald thanks for the feedback. I'm a bit disappointed by the performance of lapsolver in this respect. Did you run lapsolver also with int32 version, I does have some type...

@kylemcdonald hope you are doing well! I recently stumbled upon [IsoMatch](https://gfx.cs.princeton.edu/pubs/Fried_2015_ICI/FriedDiVerdiHalberSizikovaFinkelstein_Eurographics2015_LowRes.pdf), an algorithm to arrange a collection of objects to a target layout while trying to preserve distances between objects...