single GPU in training
chiran7 opened this issue · comments
Dear Authors,
In the training, it is mentioned that;
We use multiple GPUs for training. {YOLOv4-P5, YOLOv4-P6, YOLOv4-P7} use input resolution {896, 1280, 1536} for training respectively.
yolov4-p5
python -m torch.distributed.launch --nproc_per_node 4 train.py --batch-size 64 --img 896 896 --data coco.yaml --cfg yolov4-p5.yaml --weights '' --sync-bn --device 0,1,2,3 --name yolov4-p5
python -m torch.distributed.launch --nproc_per_node 4 train.py --batch-size 64 --img 896 896 --data coco.yaml --cfg yolov4-p5.yaml --weights 'runs/exp0_yolov4-p5/weights/last_298.pt' --sync-bn --device 0,1,2,3 --name yolov4-p5-tune --hyp 'data/hyp.finetune.yaml' --epochs 450 --resume
Can it be trained using a single GPU? If so, do we need to do any modifications in the train.py?
Thank you for your time and consideration.
Use the following command
python train.py --batch-size 64 --img 896 896 --data coco.yaml --cfg yolov4-p5.yaml --weights '' --device 0 --name yolov4-p5
Change the batch size and image dimensions as per your GPU configuration. For a GPU with 12 gigs RAM, i'm using batch size 32 and image size 320(Low resolution will give poorer result).
Can you explain the different layers? For instance, in [-1, 1, BottleneckCSP, [64]], it takes only one argument 64, but BottleneckCSP requires two inputs as arguments inside common.py.
Also, if we want to train for custom data, will it be ok if nc =80 is replaced, or do we need to change filter size as well before yolo layers in scaled yolov4?
Thank you for your time and consideration.
Use the following command
python train.py --batch-size 64 --img 896 896 --data coco.yaml --cfg yolov4-p5.yaml --weights '' --device 0 --name yolov4-p5Change the batch size and image dimensions as per your GPU configuration. For a GPU with 12 gigs RAM, i'm using batch size 32 and image size 320(Low resolution will give poorer result).
Can you explain the different layers? For instance, in [-1, 1, BottleneckCSP, [64]], it takes only one argument 64, but BottleneckCSP requires two inputs as arguments inside common.py.
Also, if we want to train for custom data, will it be ok if nc =80 is replaced, or do we need to change filter size as well before yolo layers in scaled yolov4?
Thank you for your time and consideration.