append faster-coco-eval as base backend for COCODetection
https://github.com/MiXaiLL76/faster_coco_eval
Motivation
I often use this project, but I saw it abandoned and without a public repository on github. Also, part of the project remained unfinished for a long time. I implemented some of the author's ideas and decided to make the results publicly available.
Modification
This implementation greatly speeds up the evaluation time for coco's AP metrics, especially when dealing with a high number of instances in an image.
Use cases
COCODetection(..., faster=True)
| Type | COCOeval | COCOeval_faster | Profit |
|---|---|---|---|
| bbox | 18.477 sec. | 7.345 sec. | 2.5x faster |
| segm | 29.819 sec. | 15.840 sec. | 2x faster |
I don’t know what’s going on in the unit tests, they fail outside of my changes, on other tests)
@zhouzaida @C1rN09 cannot run last test with GPU
Job was rejected because resource class gpu.nvidia.small, image ubuntu-2004-cuda-11.2:202103-01 is not a valid resource class
mb https://discuss.circleci.com/t/linux-cuda-deprecation-and-image-policy/48568
mb we need update ci for this build? CUDA 11.2 | Removed from linux-cuda-11