上传者: u011094454
|
上传时间: 2021-10-26 00:00:37
|
文件大小: 10.21MB
|
文件类型: -
pytorch yolov3 backbone
This project use threshold=0.1 for faster evaluation,while the original implementation use 0.01.
Adjust the training schedules(total epochs,lr scheduler) may further boost the performance. I pick StepLR instead of ConsinLR to accelerate training procedure. Continue training may give better results.