train
ahuxjz opened this issue · comments
COVER's training data set is only 75 pieces. Is it training together with other training sets or training alone? If it is only training COVER, how many generations does such a small data set train?
Is the f1 score calculated directly? During the test, what is the iou setting? How to understand the best output f1?
@ahuxjz
Hi, For COVER, we only fine-tuned a small number of iterations. As mentioned in the paper, the best F1 score is obtained by varying the threshold during test.