Yolov3-spp performs better on the Coco dataset than Yolov3.
When training with my on images and classes, do you recommend using the Yolov3-spp architecture as well? If so, can I use darknet53.conv.74 as inital weights for the backbone?
(So the command line would be something like ./darknet detector train cfg/my.data cfg/yolov3-spp.cfg darknet53.conv.74)
Hi @AlexeyAB,
Yolov3-spp performs better on the Coco dataset than Yolov3.
When training with my on images and classes, do you recommend using the Yolov3-spp architecture as well? If so, can I use
darknet53.conv.74
as inital weights for the backbone?(So the command line would be something like
./darknet detector train cfg/my.data cfg/yolov3-spp.cfg darknet53.conv.74
)Thank you very much!