Hello, I am new in YOLO. I have 10 pictures and size like 3023x3023,521x528,2160x2175 and 2075x2742. we need detect the thousand of objects in my single pic, I split the pic to 192x192, Because It's easy for labeling. So I got the 300 pic by resize.
If I resize my training data to 192x192, then my testing data need to be resized or can used original size? I used split pic it good accuracy, but something in the margin it can't detect. It's YOLOv3 can deal with overlaping object case? or any other algorithm can deal with?
The other qustion, I has trained the model by 192x192 pics . I load the weights to testing my original size pic, it can't detect the object. How can I tune the cfg? Or when I prepare training data don't split the pic to small size?
wahid18benz
commented
5 years ago
Hello, I am new in YOLO. I have 10 pictures and size like 3023x3023,521x528,2160x2175 and 2075x2742. we need detect the thousand of objects in my single pic, I split the pic to 192x192, Because It's easy for labeling. So I got the 300 pic by resize.
If I resize my training data to 192x192, then my testing data need to be resized or can used original size? I used split pic it good accuracy, but something in the margin it can't detect. It's YOLOv3 can deal with overlaping object case? or any other algorithm can deal with?
The other qustion, I has trained the model by 192x192 pics . I load the weights to testing my original size pic, it can't detect the object. How can I tune the cfg? Or when I prepare training data don't split the pic to small size?