Maskrcnn-benchmark: How to draw a precision-recall curve when testing.
❓ Questions and Help
question
guanbin1994
>All comments
It will require using a few internal features from COCOeval (if you are evaluating on COCO, the answer changes if not).
After this part https://github.com/facebookresearch/maskrcnn-benchmark/blob/d28845e112de36781b2b5f7217a34b2b62de8d2f/maskrcnn_benchmark/data/datasets/evaluation/coco/coco_eval.py#L289
you can grab the coco_eval object and you'll find in its .eval field both the precision and the recall, see https://github.com/cocodataset/cocoapi/blob/ed842bffd41f6ff38707c4f0968d2cfd91088688/PythonAPI/pycocotools/cocoeval.py#L412-L419
fmassa
on 18 Jan 2019
❤5
👍1
Was this page helpful?
0 / 5 - 0 ratings
Related issues
krumo
·
3Comments
Nacho114
·
4Comments
SkeletonOne
·
3Comments
qijiezhao
·
3Comments
mrteera
·
3Comments
Most helpful comment
It will require using a few internal features from COCOeval (if you are evaluating on COCO, the answer changes if not).
After this part https://github.com/facebookresearch/maskrcnn-benchmark/blob/d28845e112de36781b2b5f7217a34b2b62de8d2f/maskrcnn_benchmark/data/datasets/evaluation/coco/coco_eval.py#L289
you can grab the
coco_evalobject and you'll find in its.evalfield both theprecisionand therecall, see https://github.com/cocodataset/cocoapi/blob/ed842bffd41f6ff38707c4f0968d2cfd91088688/PythonAPI/pycocotools/cocoeval.py#L412-L419