Closed yuheng222 closed 4 years ago
Solved by downgrading torch from 1.4.0 to 1.3.1, torchvision from 0.5.0 to 0.4.2, pillow from 7.0.0 to 6.2.1.
I think this was caused by torch1.4.0 and its compatibility with PIL image.
Yes, I think so too. Thanks for the reply!
Just another question, if I would like to train my model with multiclass dataset, which lines do I have to change? I tried changing all the sigmoid in pytorch_run.py to softmax but it didnt work since the dice score is still low. Thanks.
Did you change the input and output for the model? And by changing the to softmax it should work. What's the dice score?
The dice score is 0.23958333333333334.
For the input shape, I tried changing the line: model_test = model_unet(model_Inputs[0], 3, 1) to model_test = model_unet(model_Inputs[0], 3, 8) since my there are 8 classes in my dataset labels but I get this error: ValueError: Target size (torch.Size([4, 1, 96, 96])) must be the same as input size (torch.Size([4, 8, 96, 96])). Where can I change the shape of the model output? I can't seem to find the line to change output shape in pytorch_run.py. Thank you for helping!
Dice score = 0.24 is not good at all. Yes by changing it to (model_input[0],3,8 ) it should be working for 8 classes.
Also, you have to change it to softmax to make it work for all prediction line. pred_tb = F.sigmoid(pred_tb)
Can you tell me which line you are facing this issue?
Hi there,
I'm relatively new to medical image segmentation and I want to use your models to experiment on some CT/MRI images and labels in NIFTI (.nii.gz) format.
I converted the 3D NIFTI images and labels to 2D images in .png format using your 2d_from_3rd.py script, but after running pytorch_run.py, I was faced with the following error:
May I know what might be the cause of this error? Thanks!