Hoel KERVADEC

I suppose you would need to add another elif [in there](https://github.com/SublimeText/LaTeXTools/blob/master/builders/basicBuilder.py#L151)? ```python if CITATIONS_REGEX.search(self.out): run_bibtex = True # are we using biblatex? m = BIBLATEX_REGEX.search(self.out) if m: bibtex = m.group(1).lower()...

I guess you could launch Tilda from another terminal ; this way you would have some debugging informations when it crashes.

Alas I am not allowed to share that. You have to create an account and apply yourself. Note that in the meantime, a v2 of the dataset (with more annotated...

Hello, The one to use is [`NaivePenalty`](https://github.com/LIVIAETS/SizeLoss_WSS/blob/master/losses.py#L52) for the 2D case, and `BatchNaivePenalty` that was used for the 3D setting. In our case the bounds were computed at the dataloader...

Hi, I don't think you have done anything wrong, but rather that this codebase is due for an update (2 years old), as the latest pytorch versions have introduced some...

The solution is to pre-compute offline the distance map in 3D, save them into a `.npy` in the with the axises `kxyz`, with `k` being the class axis. Pay attention...

With respect to the data augmentation > there is no better choice other than computing the SDM on the fly When you refer to "on the fly", you mean to...

> Thanks for your response. I tried your loss function on our dataset, however, I have not seen improved performance till now. I want to know if the simplification from...

That one fell through the cracks (sorry for that), please feel free to re-open/reply if still relevant.

Hi, and thanks for the interest in our work. I'm trying give you a partial reply now, but I might come back in a few days once I've had more...