I find some bugs in this file which is named by Detectron/detectron/datasets/voc_eval.py.
From line 213 on:
compute precision recall
fp = np.cumsum(fp)
tp = np.cumsum(tp)
rec = tp / float(npos)
# avoid divide by zero in case the first detection matches a difficult
# ground truth
prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps)
ap = voc_ap(rec, prec, use_07_metric)
we can see that the rec and prec is not a list but a float number.This means the calculation of ap is not the area of PR curve.
So could you please tell me how to fix or make sense of this question.
I find some bugs in this file which is named by Detectron/detectron/datasets/voc_eval.py. From line 213 on:
compute precision recall
we can see that the rec and prec is not a list but a float number.This means the calculation of ap is not the area of PR curve.
So could you please tell me how to fix or make sense of this question.