Kerem Turgutlu

Could you share which notebook and fastai + pytorch versions you ran with. Sorry for replying this late :D If this is no longer a problem of interest I can...

Hi, Thanks for heads up, working on it right now ! Best

In `EDA - PART 2 CONTOUR LABELS` notebook you may see the pipeline and functions for contoured mask target creation with visual plots. Hope this helps

Currently not, but it would definitely be a good addition, happy to take PRs 😄 Good part is DICOM is a very well defined data structure and it should be...

Which python version are you using? It may not be supporting f strings. Use python 3.6+

I am glad it helped :) Basically this information comes from dicom metadata which stores `x,y spacings`, `origin` and `x,y positions in mm`. This equations converts cartesian coordinates which is...

Thanks for pointing this out! Please feel free to do PRs :)

Great discussion! I am a bit confused regarding this part from the paper: ``` The calculation of embedding similarities was also sharded with individual GPUs computing only the subset of...

> @KeremTurgutlu I found the same question when I want to train CLIP with 8 GPUs. I found that if my total batch size is 128, then pytorch will split...

> I am curious if you or Kerem have successfully tried any of the tricks mentioned above to increase the effective size of the contrastive pairs. So far, I tried...