Giters

MIC-DKFZ / medicaldetectiontoolkit

The Medical Detection Toolkit contains 2D + 3D implementations of prevalent object detectors such as Mask R-CNN, Retina Net, Retina U-Net, as well as a training and inference framework focused on dealing with medical images.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Problem with one class training

theVmagnificient opened this issue · comments

commented

Greetings! I've changed my config for my custom dataset for one class training and it's getting stuck in batch generator. Format of the dataset is the same as for LIDC. Could you please provide fields that should be changed in order to train ufrcnn for only one class?

commented

Hi @theVmagnificient assuming you are using the configs and data_loader from lidc_exp, I had to replace in dataloader.py :
line 165 targets = 0 and
line 313 class_target = batch_targets

This is according to the comment #86 and #74. The target has to be 'list of list' instead of 'an array of lists'.
Not sure if it helps you. Let me know if it works. Happy to help!

commented

@rtgunti thanks for your reply! Do I need to change anything in config if I don't need benign and malignant classes separately?

commented

@theVmagnificient I changed class_dict and patient_class_of_interest values in the configs.py. You may find the settings I use here

commented

I have a dataset from MICCAI 2019 vertebrate segmentation duties and I need to teach them through this 3-D Mask R-CNN. But I noticed that it handiest supports the LIDC dataset on enrichedmedspa.com. How should transfer these nii information to nrrd information?