register dataset in COCO format #3056
andresviana
started this conversation in
General
Replies: 2 comments
-
I don't know how are you handling your paths, but it seems you have a problem with different systems (windows and Linux) ways to handle the path. In python a good politics is to use os.path.join(firstPart, secondPart,....) (https://docs.python.org/3/library/os.path.html) so that you are independent of the architecture you are using. I hope it helps |
Beta Was this translation helpful? Give feedback.
0 replies
-
Thanks Daniel for your answer!! I think that issue it's related in how DatasetCatalog.register handling the paths when it creates Metadata dictionary from Google Colab. I'll be trying. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi guys, in colab I'm following the tutorial for register a custom dataset on COCO annotations format:
from detectron2.data.datasets import register_coco_instances
register_coco_instances("my_dataset_train", {}, "./content/train/annotations.json", "./content/train/images")
register_coco_instances("my_dataset_val", {}, "./content/val/annotations.json", "./content/val/images")
register_coco_instances("my_dataset_test", {}, "./content/test/annotations.json", "./content/test/images")
I found in Standard Dataset Dicts that file_name charged the full path to the image file but with a double back slash before the image file name:
'file_name': './content/train/images/JPEGImages\\frame5127.jpg'
Any idea about how could I fix it?
Thanks!!
Beta Was this translation helpful? Give feedback.
All reactions