Soumya Snigdha Kundu

Could you please implement a feature where we can choose the ratio for how much sampling is required. I believe now, it automatically leads to equal weight. Something like this:...

Remove rng parameter of binary_blobs() function in plot_random_walker_segmentation.py

8

Using the rng parameter throws the following TypeError: TypeError: binary_blobs() got an unexpected keyword argument 'rng' ## Description Parameter rng removal in line: `data = skimage.img_as_float(binary_blobs(length=128, rng=1))` of Random walker...

I think 32 is a better start. Especially for those who want to test it on the free tier colab.

Potential Unsupported version of DICE.

5

### Bug description https://lightning.ai/docs/torchmetrics/stable/classification/dice.html Here, under average parameter of the dice score, there is the option of 'weighted' mentioned. But that triggers an error while using it. ### How to...

Some issue with loading model with 'weight' as opposed to 'pretrained=True'

1

1. `net = resnet18(weights=ResNet18_Weights.IMAGENET1K_V1)` #new pytorch style 2. `net = resnet18(pretrained=True)` #opacus tutorial/old style Would there be any reason for the first one to fail? I get an error while...

Better augmentation example in starter notebook.

1

### 💡 Your Question In the [demo notebook](https://colab.research.google.com/drive/1q0RmeVRzLwRXW-h9dPFSOchwJkThUy6d): There are simply 2 cells which go: > SuperGradients added transforms for you. You're free to experiment with these transformations as you...

20210418 Bingyuan Liu The hidden label-marginal biases of segmentation losses [(pytorch)](https://github.com/by-liu/SegLossBias) [arxiv](https://arxiv.org/abs/2104.08717) This is already there as a media paper, maybe remove the arxiv version? or just link the paper?