Closed YuQi9797 closed 3 years ago
Hi @YuQi9797, please refer to the accuracy_2animal() for more details. Basically, the distance is firstly normalized by 'output.size(3)/10' and them compared to a threshold 0.5. That is how the 0.05 comes from. We adopt this evaluation metric from previous work. Hope this clarifies.
Sir, Is there no difference between accuracy_2animal () and accuracy ()?
The only difference is that the accuray () is used to train and calculate the 19 key points, while the accuracy_animal () is used to calculate the 18 key points. But essentially there is no difference between the two functions.
I see the code in https://github.com/chaneyddtt/UDA-Animal-Pose/blob/f1ebfda860/evaluation/test.py#L140 , there are not threshold about 0.05. and the
pck_score[j] += acc[j + 1].numpy()
calculates acc instead of PCK ?Can someone tell me where I'm going wrong?