I have been trying for about 6 hours to figure out _annotations.coco.json or any other file that will work for labeling images in bulk. There is no way to trouble shoot this. First nothing worked and then I rearranged the sections and now everything is a label but truncated. There is no good documentation to describe how to bulk label. There are no error messages when the file is incorrect and no way to validate what correct looks like.
In addition; there is no way to delete all of the classes if weird import does occur. It’s delete the project and start over.
I have about 59,000 images I want to run through the model or perhaps segment it.
There needs to be a better primer for annotations and some online tools for checking the files.
Hey @cbitterfield - sorry for your trouble here!
Do you mind walking me through the workflow you’d like to see, step-by-step? Are you trying to upload already-labeled images? How did you create those annotation files?
@Jacob_Witt I have the same problem. It is unclear from What is the COCO JSON Annotation Format? where the annotation file should be relative to the images (I’m going to assume they should be in the same directory), and if there is any requirement what the annotation file should be called. I have tried to upload a minimal example using the coco.json from the above URL, and a random image.
test/
- coco.json
- 0001.jpg
I uploaded this dataset as
roboflow import -w myworkspace -p myproject ./test
Afterwards, I could see 0001.jpg in the project under unassigned, but there is no indication that roboflow ever processed the annotation file, or even received it. This is an instance segmentation project and I uploaded from Windows 10 in Git Bash.