def eval_ap(iou, iou_idx, cls, gt, predition):
ap = evaluate.ap(predition, iou[0], gt)
sys.stdout.flush()
return cls, iou_idx, ap
for cls in prediction_by_cls:
all_prediction += cls
print("NMS performed.")
# calculate
ious = np.arange(0.1, 1.0, 0.1)
aps = np.zeros((len(ious), label_count))
ars = np.zeros((len(ious), label_count))
f1s = np.zeros((len(ious), ))
miou = evaluate.miou_per_v(all_prediction, all_groundtruth)
for i, iou in enumerate(ious):
for cls in range(label_count):
ap, ar = evaluate.ap(prediction_by_cls[cls], iou,
groundtruth_by_cls[cls])
aps[i][cls] = ap
ars[i][cls] = ar
f1 = evaluate.f1(all_prediction, iou, all_groundtruth)
f1s[i] = f1
map_ = np.mean(aps, axis=1)
mar = np.mean(ars, axis=1)
print("Criteria solved.")
# print
title = "C3D Detection Performance"
datas = [["IoU threshold"], ["mean AP"], ["mean AR"], ["F1 criterion"]]
for i, iou in enumerate(ious):
datas[0].append("{:.2f}".format(iou))