I'm trying to understand this code with precision vs recall. I'm confused as to why the recThrs collection is not used here for recall - it seems like the iouThrs buckets are used instead - see line 450 below. Is this a copy/paste bug?
I understand recall to be "count the number of detections that matched the ground truth above threshold X and then average that in the range. The code seems to be averaging that against the 10 buckets of iouThrs (0.5 - 0.95 inclusive step 0.05). The recThrs is defined in the code as (0-1 step 0.01). Shouldn't it be using recThrs?
Hi,
I'm trying to understand this code with precision vs recall. I'm confused as to why the recThrs collection is not used here for recall - it seems like the iouThrs buckets are used instead - see line 450 below. Is this a copy/paste bug?
I understand recall to be "count the number of detections that matched the ground truth above threshold X and then average that in the range. The code seems to be averaging that against the 10 buckets of iouThrs (0.5 - 0.95 inclusive step 0.05). The recThrs is defined in the code as (0-1 step 0.01). Shouldn't it be using recThrs?
Here's the code in
https://github.com/cocodataset/cocoapi/blob/bab68f13221afca2d053c877806886d7e2ce9dde/PythonAPI/pycocotools/cocoeval.py#L438-L452