A simple and extensible object detection evaluator in COCO style
Features
- Extensible user interfaces to deal with different data formats.
- Support customized evaluation breakdowns, e.g., object size in COCO, difficulty in KITTI, velocity and range in Waymo.
- Interface for general matching scores, e.g. 2D IoU, 3D rotated IoU, center distance.
- Support widely-used metrics like AP, AR and customized metrics like average regression errors, average IoUs, etc.
- Purely based on Python, easy to develop your customized metrics.
Installation
pip install treelib- Clone this repo and run
pip install .in the cloned directory, orpip install -e .if you want to modify the source code.
Prepare predictions and groundtruth
You need to define a function to read the predictions and groundtruth.
def read_prediction(path):
...
return results
where