ZFTurbo / Keras-RetinaNet-for-Open-Images-Challenge-2018

Code for 15th place in Kaggle Google AI Open Images - Object Detection Track
MIT License
268 stars 76 forks source link

Hi, how to use multi gpus? #13

Open louielu1027 opened 5 years ago

louielu1027 commented 5 years ago

I use 8 gpus for infenrence, but it is slowly than 1 gpu...... I changed 'gpu_use = 0' to 'gpu_use = 0,1,2,3,4,5,6,7' in retinanet_inference_example.py. Is it right?

ZFTurbo commented 5 years ago

No, this variable only controls which GPU use for calculation. Code is for single GPU only. If you need inference on several GPU, just split data in N parts and run inference in parallel on each of them.

louielu1027 commented 5 years ago

Ok, Thanks!