No project description provided
This package wraps a facebook C++ implementation of COCO-eval operations found in the pycocotools package. 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.
For our use case with a test dataset of 1500 images that contains up to 2000 instances per image we saw up to a 100x faster evaluation using fast-coco-eval (FCE) compared to the original pycocotools code.
Seg eval pycocotools 4 hours Seg eval FCE: 2.5 min BBox eval pycocotools: 4 hours BBox eval FCE: 2 min
pip install fast-coco-eval
If you clone the repo and install it locally, the following command is recommended
pip install -e .
given that you are in the fast-coco-eval directory. There seem to be an issue with loading the C++ extensions when installing it from the root directory without the -e flag.
This package contains a faster implementation of the
Due to torch being used to compile and access the C++ code, it needs to be imported before using the package.
To import and use
import torch from fast_coco_eval import COCOeval_fast
For usage, look at the original
COCOEval class documentation.
It would be nice to decouple it from the pytorch build tool for the c++ compilation.
- Wrap c++ code
- Get it to compile
- Add COCOEval class wraper
- Remove detectron2 dependencies
- Check if it works on windows
- Remove torch dependencies
Distributed under the apache version 2.0 license, see license for more information. © 2021 Sartorius AG
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.