WongKinYiu / ScaledYOLOv4

Scaled-YOLOv4: Scaling Cross Stage Partial Network
GNU General Public License v3.0
2.02k stars 572 forks source link

calculating the performance #251

Open mrlmnq opened 3 years ago

mrlmnq commented 3 years ago

in Method called ap_per_class https://github.com/WongKinYiu/ScaledYOLOv4/blob/yolov4-large/utils/general.py#L220-L280

fbc is the FP tpc is the Tp

is the n_p FN n_gt TN

so I can calculate the accuracy

acc = tpc + n_gt /fbc+tpc+n_gt+n_p

is this right also why

recall = tpc / (n_gt + 1e-16) # recall curve 1e-16 indicate to ?

mrlmnq commented 3 years ago

@WongKinYiu

WongKinYiu commented 3 years ago

no, n_gt is number of ground truth. 1e-16 is used to avoid divided by zero.

mrlmnq commented 3 years ago

ty @WongKinYiu where is the False-negative (FN) and True-negative (TN)

mrlmnq commented 3 years ago

@WongKinYiu

mrlmnq commented 3 years ago

anyone here can assist :)

WongKinYiu commented 3 years ago

for AP50 case

https://towardsdatascience.com/evaluating-performance-of-an-object-detection-model-137a349c517b