The size of the picture used for training

qqwweee / keras-yolo3

A Keras implementation of YOLOv3 (Tensorflow backend)

MIT License

7.13k stars 3.45k forks source link

The size of the picture used for training #672

Open tangbinbinyes opened 4 years ago

tangbinbinyes commented 4 years ago

Hi, do all pictures used for training have to be the same size? If the size is different, does it have an impact on the generation of anchor and the prediction effect of the model? The pictures I used to train here are all 960 480, but later I added a lot of pictures with a size of 1920 1080. I do n’t know if the inconsistent picture size will have a bad effect on the detection effect of the model , Looking forward to your answer very much, I would appreciate it!

IPENCIL666 commented 4 years ago

i have same problem!

robisen1 commented 4 years ago

The answer is no but if you use really large images you may find problems with memory.