VNext
VNext copied to clipboard
Training/Validating on Custom Datasets (In Classic COCO Segmentation Format)
Hi all,
Does this repo support training on a custom dataset? Or better still, has anyone successfully done so? I have a dataset in classic coco format which I am able to load into detectron2 with register_coco_instances. This dataset is one that contains a series of annotated images that can be combined to form a video. Currently, I am applying plain image instance segmentation using detectron2 PointRend but IDOL seems very promising to me.
In the case that this has not been done before, I would appreciate some pointers that would guide me toward writing some functionality on top of this repo to implement training/validating on a custom dataset.
Thanks and congrats on your paper!
I am currently facing similar thing, I want to train custom dataset, but i don't know how?