aim-qmul / sdx23-aimless

Source Separation training codebase for the Sound Demixing Challenge 2023.
35 stars 0 forks source link

feat: make data augmentation pipeline configurable #26

Closed yoyololicon closed 1 year ago

yoyololicon commented 1 year ago

Dropping apply_transforms from LightningModule and LightningDataModule. Adding the transforms argument, which accepts a list of transformation objects which can be specified in yaml. For LightningModule, the transformation objects should inherit augment.cuda.CudaBase, while for LightningDataModule it's augment.cpu.CPUBase.

close #22