NanoDet-Plusā”Super fast and lightweight anchor-free object detection model. š„Only 980 KB(int8) / 1.8MB (fp16) and run 97FPS on cellphoneš„
5.78k
stars
1.04k
forks
source link
Traning code freezed when saving best check point #539
Open
aemior opened 1 year ago
I use 4x4090GPU and mobileone backbone with batch size 50 to train nanodet, the memory per gpu is about 23.3GB, but the training code is freeze when save the best check point at this line: https://github.com/RangiLyu/nanodet/blob/4d85d0cb15aa076261a85ab93e7d9e45b7f64fda/nanodet/trainer/task.py#L273 Any ideal to debug?