COCO Detection Challenge (Segmentation Mask) Forum

Go back to competition Back to thread list Post in this thread

> Submissions Failing


I have submitted multiple JSON files for the COCO segmentation masks competition which have been successful (100+ submissions) -

However new submissions (after 19th July 2023) are now all failing giving the error below.

WARNING: Your kernel does not support swap limit capabilities or the cgroup is not mounted. Memory limited without swap.
/opt/conda/lib/python2.7/site-packages/matplotlib/ UserWarning: Matplotlib is building the font cache using fc-list. This may take a moment.
warnings.warn('Matplotlib is building the font cache using fc-list. This may take a moment.')

I am using the mmdetection framework and training with a model that outputs both object detection and instance segmentation results. The object detection (bounding boxes) submissions are working fine, however, the instance segmentation submissions are failing.

Downloading the output file and looking at the metafile for a failed submission gives the below;

elapsedTime: 268.6641709804535
exitCode: 137
ingestionExitCode: 0

Posted by: Cathaoir @ July 27, 2023, 2:17 p.m.

What I've found from playing around with submissions and looking into the forums from the old server for the bounding box competitions.

The above error is due to the submitted JSON file being too large. From submissions, it appears once the original (non-zipped) JSON is approximately over 1.06 GB, the submission will fail.

Ways to reduce JSON file size. As stated I have used mmdetection so there may be differences in frameworks and how they generate the submission JSON file.

- Remove the bbox key from the JSON, this is not needed since this is the segmentation mask competition.
- Reduce the decimal points to 2/3 decimals places instead of mmdetections default 17.

I hope this helps!

Posted by: Cathaoir @ July 28, 2023, 5:40 p.m.
Post in this thread