MIScnn
MIScnn copied to clipboard
A framework for Medical Image Segmentation with Convolutional Neural Networks and Deep Learning
Hi, when training a model on multiple GPUs with patch-wise cropping, I get the following error: batches.pop(0) from empty list I train with a patch size of (64,64,64), 4 Gpus...
Hi again, one more thing I noticed (sorryyyy): When you use patchwise_grid and set the prepare batches flag, I think that the number of steps per epoch is not updated...
Hi, I just noticed that if you set the 'prepare-batches' flag while training with patch-wise cropping, the network uses the same patches for each epoch instead of creating a new...
**Currently sitatuion:** Preprocessed data sets via subfunctions or complete prepared batches are stored in the batches directory with a unique seed. But, there is no re-use possibility of these. **Desired...
**When does it occur:** If the number of samples is very low and the batch size is higher than the number of available samples. Can especially occur in full-image analysis...
@muellerdo A new and high promising API has been emerging in `Keras`, called [`KerasCV`](https://github.com/keras-team/keras-cv). It would be great if frankkramer-lab /MISCnn consider to contribute some medical imaging component.
The patch combination after prediction sometimes works, sometimes do not in the current dev branch. Bug searching required! something todo with even/uneven and in combination with padding?
Update NiftiSlicerIO to utilize sample instead of pred and index
When `gt` and `pd` are exactly the same and filled with the same value `val`, [compute_dice](https://github.com/frankkramer-lab/MIScnn/blob/master/miscnn/evaluation/detailed_validation.py#L72) returns NaN for the classes that are not the value `val`. This is due...
``` ct_scans/ scan_a/ segmentation.nii.gz imaging.nii.gz scan_b/ segmentation.nii.gz imaging.nii.gz scan_c/ segmentation.nii.gz imaging.nii.gz ``` Above is the file structure of my dataset where segmentation.nii.gz and imaging.nii.gz are 3D NIFTI files with multiple...