milesial comments

Results 62 comments of


                                            milesial

Why do you need to fill 0 when convolution operation？

The image size decreases when there is no padding. Adding padding makes it so that the image size stays the same across all layers that are on the same level....

can we use this to repeat arxiv.org/abs/1505.04597

Yes you can, it's mainly a matter of correctly preprocessing the MRI data. The building blocks of the UNet are in this repo.

n_classes and Target size

`n_classes=2` with 0/1 targets should work. Where do you see the black images, in wandb or when running the predict script? Also, how long did you train the network for?

When I change the dataset, the code of 'loss = criterion(masks_pred, true_masks)' of train.py put error

The target class 255 is bigger than the number of classes you specified for the network, you should fix your mask preprocessing.

mask format problem

Your mask should be boolean: 0/1. Not 0/2, not 0/255. You say that it's training fine with a boolean mask. What is the issue then?

TypeError: evaluate() takes 0 positional arguments but 3 were given?

Can you provide a full traceback?

Using U-Net for Real-time semantic segmentation

Hi, I think that 10 images is very low. Even with data augmentation, you don't have much information. I would recommend collecting more images (in the hundreds) and applying data...

Dice coefficient no change during training,is always very close to 0

Yes the mask images should be B&W, not RGB. Try that.

Dice coefficient no change during training,is always very close to 0

if the loss is negative you are loading your masks wrong, check that you correctly modified the preprocess method

Dice coefficient no change during training,is always very close to 0

@Cassie0207 You have to check that your loaded masks after the preprocess method, if you have 2 or 1 classes, are 0s and 1s only. If they are not, you...