Open forever208 opened 3 years ago
I changed the conf_thresh
to 0.5
in the do_detect()
call, so this time it won't generate several thousands of bounding boxes from do_detect()
function. But the evaluation only does batch_size=1 evalution which seems quite slow for all the images, so I think I'm gonna change that to 16 or 32.
too slow for searching the image labels in JSON file, around 10 minutes for each image.
by the way, there are 2 bugs in
evaluate_on_coco.py
(one is the data type of the image, another is the argument number error )
how to fix bro
I changed the
conf_thresh
to0.5
in thedo_detect()
call, so this time it won't generate several thousands of bounding boxes fromdo_detect()
function. But the evaluation only does batch_size=1 evalution which seems quite slow for all the images, so I think I'm gonna change that to 16 or 32.
change where bro
too slow for searching the image labels in JSON file, around 10 minutes for each image.
by the way, there are 2 bugs in
evaluate_on_coco.py
(one is the data type of the image, another is the argument number error )