Dear author and clever workers,Has anyone used PETR to train your scenarios and data? I used petrv1 to train my own labelled fence(long and thin,along the road) and digger( small and medium) data,including 186 sets of fence and 359 sets of digger data.The validation set was tested with a mAP of 0 and a target score of no more than 0.1,with a large deviation in both size and position. Is it that PETR is not applicable to my data or the amount of my data is too small, maybe I need to make network structure adjustments and where should I start?What should I do?
Dear author and clever workers,Has anyone used PETR to train your scenarios and data? I used petrv1 to train my own labelled fence(long and thin,along the road) and digger( small and medium) data,including 186 sets of fence and 359 sets of digger data.The validation set was tested with a mAP of 0 and a target score of no more than 0.1,with a large deviation in both size and position. Is it that PETR is not applicable to my data or the amount of my data is too small, maybe I need to make network structure adjustments and where should I start?What should I do?