Question about correlation between real localization quality and top-1 generated distribution at training time
pjh4993 opened this issue · comments
Does correlation between real localization quality and top-1 value of general distribution are consistent during training? I'm wonder that those relations are consistent during whole training time not just after traning.
I have the same question,too. Since I need to estimate the bbox localization quality during the training.Maybe I should have a test?I don't know.
I've accomplished an experiment, in the beginning, they're not very correlated (R square about 0.5 in my dataset). After some time of training, it goes to 0.8 and higher. This feature really benefits my research project, I'll discuss this more in my project or paper in the future.
I tested the correlation between iou(target and pred) and top-1 distribution. Maybe you can choose other metrics.