As you know, this challenge has a timeout of 1 second per image.
But we have been notified by multiple participants that their submission getting timeouts on partial datasets or a few images and so on.
We have done the following changes to make the timeouts a bit relaxed.
1. Throughput check instead of time taken for individual images
This should help for cases when a few images take a bit longer due to reasons like different image resolution, your system’s resources utilization, or just random bad luck.
2. Soft timeouts
In case any image is taking longer than 3 seconds, we ignore that image for your submission going forward and assume no predictions are generated on it. This ensures that a few slower predictions don’t impact your throughput a lot.
3. Increase in timeout
The revised timeout for this challenge is 1.5 seconds/image.
We hope that this change helps you and you can improve your models instead of worrying too much about timeout.
All the best!