search

MASILab / 3DUX-Net

250 stars 34 forks source link

model was not saved！ current best Avg，Dice：0.0 current Avg Dice：nan #29

Closed aixiaobaikkk closed 1 year ago

aixiaobaikkk commented 1 year ago

Hi , thank you very much for your great work，I have meet some trouble when I train the amos dataset，it just shows ：

l 21/240 [01:17<11:29，3.15s/it] Model Was Not Saved ! Current Best Avg. Dice: 0.0 Current Avg.Dice: nan

I wander why it happened，how to solove it

leeh43 commented 1 year ago

Hi, thank you for your interest towards our work! Just want to have more information, have you changed any lines of code?

SanmAIGO commented 1 year ago

@leeh43 Hello, I also had this problem with the AMOS2022 dataset. I modified： RandCropByPosNegLabeld( keys=["image", "label"], label_key="label", spatial_size=(96, 96, 96), pos=1, neg=1, num_samples=crop_samples, image_key="image", image_threshold=0, allow_smaller=True #changed ), The above modification is to avoid： ValueError: The size of the proposed random crop ROI is larger than the image size.

That's why, and what can I do to replicate your experiment on AMOS2022. ValueError Error log I provided in this issue#32 Looking forward to your reply. Sanm

leeh43 commented 1 year ago

Hi @aixiaobaikkk, I have double checked the main_finetune.py code for finetuning and it didn't happen your error with the following screenshot:

image

Just want to clarify the AMOS experiment in our paper is only for CT organ segmentation, please refer the dataset json file for task 1 in AMOS challenge.

Hi @SanmAIGO, please keep the original code for the data transform and I believed the code is doing good with the above screenshot.

aixiaobaikkk commented 1 year ago

@leeh43 Thanks for your reply ，I was too busy studying during this time to continue working on the project, after which I would check my code and continue learning this great project