The question about multi-scale training and preset prior anchors config

argusswift / YOLOv4-pytorch

This is a pytorch repository of YOLOv4, attentive YOLOv4 and mobilenet YOLOv4 with PASCAL VOC and COCO

1.68k stars 330 forks source link

The question about multi-scale training and preset prior anchors config #209

Open Klutch0615 opened 3 years ago

Klutch0615 commented 3 years ago

If I use multi-scale training, do I need to adjust the "ANCHORS" in the _yolov4config.py? Because my understanding is that the parameters set here are related to the size of the input image, whether it is 416 or 224, but different input sizes(due to the multi-scale training) will cause different feature map sizes, so the ANCHORS preset on it should be adjusted. Am I right? Thanks