Christopher

Maybe instead there could be 2 types of boards: "volatile" boards that are always deleted after the last user leaves, and "persistent" boards that are preserved, like current boards. This...

Hey @HeinrichAD thanks a lot for your feedback! For clarification: Hooks are only ever active after `composite.register` and before `composite.remove` has been called, which is done at the beginning and...

Hey @HeinrichAD I'm currently working on the rest of the documentation for this, but functionality and tests are now finished (unless I find a bug or something missing). If you...

Hey @HeinrichAD I am done with the PR and would like to merge. If you have not checked already, you can see if it works for you as expected. A...

Hey @HeinrichAD thanks a lot again for your feedback. > In general, the code looks good. It also works as I expected. Only one thing is confussion me: The output...

Hey @bugsuse , I think this should simply just work if you use any of the composite rules. Since the gradient of `nn.Upsample` will just be constant depending on the...

I expect you already made sure the model runs when doing a forward pass without Zennit? (I.e., without the Attributor context) Could you supply the model you are using? There...

> Are you saying that I should supply the tensor sizes for each layer of unet model as tuple to `attributor`? No, I was only referring to whether your layers...

Do you only propagate zeroes back? You should try it the way @maxdreyer suggested, but since you only have a single class, I would try to simply use the full...

I do not really know anything about the data, but this could potentially mean the prediction for the positive mask values is very local. Alternatively you could try to set...