Do not using my GPU

[Lee-JeongMin]

I training my own dataset. But, darkflow is not using my GPU. How to use GPU? MY command is CUDA_VISIBLE_DEVICES=1 python3 flow --model ./cfg/my-yolo.cfg --labels ./labels.txt --dataset ./data/corn_img/ --annotation ./data/Result_corn_old-batch 32 --epoch 10000 --save 1000 --keep 50 --lr 1e-04 --gpu 0.9 I already tried to --gpu 1.0 but, darkflow only use CPU.

AND do you know how to use multi gpus? i have 4 gpus(0,1,2,3) but I use only 2 gpus(2,3)

How can i fix it??

[weiwei106]

I meet the same issues.

[weiwei106]

I training my own dataset. But, darkflow is not using my GPU. How to use GPU? MY command is CUDA_VISIBLE_DEVICES=1 python3 flow --model ./cfg/my-yolo.cfg --labels ./labels.txt --dataset ./data/corn_img/ --annotation ./data/Result_corn_old-batch 32 --epoch 10000 --save 1000 --keep 50 --lr 1e-04 --gpu 0.9 I already tried to --gpu 1.0 but, darkflow only use CPU.

AND do you know how to use multi gpus? i have 4 gpus(0,1,2,3) but I use only 2 gpus(2,3)

How can i fix it??

maybe you can use other tensorflow vrsion.

[Lee-JeongMin]

@weiwei106 I fixed it! I’m using tensorflow 1.6.0 It is no matter. I use docker, I add tensorflow enviroment on docker. So i use GPU, but i think can’t using multi GPU.

[yo3hcv]

I had the same issue (Windows 7, GTX1070) Basically tensorflow-gpu depends on many other packages, pip install is not sufficient.

I uninstalled everything, than let conda to manage all

conda install tensorflow-gpu

Now it works great, confirmed by GPU-Z