search

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.
Apache License 2.0
1.31k stars 297 forks source link

Iterate over all training data only once per epoch #109

Open sinitame opened 4 years ago

sinitame commented 4 years ago

Hi, I'm trying to make some tests using the lidc_exp experiement and I want to iterate over my training data only once per epoch. The actual configuration is iteration throughout a fix number of batches and each batches are picked randomly in the training dataset batch_ixs = np.random.choice(len(class_targets_list), self.batch_size). Which means that some samples can be seen multiple times and others might not be seen at all.

Is there a way to change this in order to go through the training set samples only once per epochs ?

JXQI commented 3 years ago

have you solved it?

bdzyubak-aiq commented 11 months ago

This could be implemented but is contour to the concept of augmentation. Random samples are pulled and randomly augmented a fixed total number of times in each epoch.