Hi, I read the code, well done, and some some suggestions as follow:
- an example for training ResNet50 is great.
- there are some customised augmentation methods, could we use the existing APIs such as
tl.prepro or imgaug instead?
- dataset folder can be merged into
tl.files ? and use the using API format?
- use lazyimport to import cv2, so that TL will import cv2 if only if cv2 is used.