search

MhLiao / TextBoxes_plusplus

TextBoxes++: A Single-Shot Oriented Scene Text Detector
Other
953 stars 279 forks source link

Why is the gpu utilization too low? #69

Closed wsai2018 closed 5 years ago

wsai2018 commented 6 years ago

The train_lmdb is 12G，and I use one TITAN Xp to train. I set the batch is 16, if i add the batch size , cuda will report memory error. Now the GPU utilization is lower than 10%,and the speed is too low,about 40 minutes per 100 iterations. THANKS!

MhLiao commented 6 years ago

@wsai2018 It depends on the resolution of the original images. The data loader and online data augmentation are too heavy if the resolution of the original images is too large.