Augmentations of detection model

[Body123]

I want to ask how I can augment a small dataset for a detection model in DocTR system (if there is a tutorial available augmentation code).

[felixdittrich92]

Hi @Body123 👋,

We apply already by default several augmentations on the fly:

val:

doctr/references/detection/train_pytorch.py

Line 247 in 9e083c4

sample_transforms=T.SampleCompose(

train:

doctr/references/detection/train_pytorch.py

Line 319 in 9e083c4

img_transforms = T.OneOf([

Feel free to adjust this to your own needs - the Resize needs to be kept as is.

Best regards,
Felix