PixelLib
PixelLib copied to clipboard
Train with a custom COCO dataset?
Is there a tutorial or documentation of how to feed a COCO dataset for custom training?
The tutorial I see uses a LabelMe service and the dataset has a JSON annotation for every image.
The only example I've found on how to load the dataset is train_maskrcnn.load_dataset("Nature")
. What if I have a custom dataset with a COCO annotation?
I am using COCO format, so I only have a single COCO JSON for the whole dataset. How would I feed this COCO dataset into the custom training?
Any advice is appreciated!