search

WongKinYiu / ScaledYOLOv4

Scaled-YOLOv4: Scaling Cross Stage Partial Network
GNU General Public License v3.0
2.02k stars 575 forks source link

Multi-GPU Training for Scaled-YOLOv4 #179

Closed xjinai closed 3 years ago

xjinai commented 3 years ago

@WongKinYiu: Thank you for your implementation of Scaled-YOLOv4. Great job!

Does PyTorch version of Scaled-YOLOv4 implementation support multi-GPU training? How is the training speed comparing with the Darknet implementation? Thanks in advance.

xjinai commented 3 years ago

For anyone interested, I found the example use of multi-GPU training as following in #109: python -m torch.distributed.launch --nproc_per_node 4 train.py --batch-size 16 --img 1536 1536 --data '../data.yaml' --cfg ./models/yolov4-p7.yaml --weights '' --sync-bn --device 0,1,2,3 --name yolov4-p7

The open question is "How is the training speed comparing with the Darknet implementation?" if anyone has experience with both PyTorch and Darknet versions.