TuSimple / TuSimple-DUC

Understanding Convolution for Semantic Segmentation
https://arxiv.org/abs/1702.08502
Apache License 2.0
610 stars 118 forks source link

How much GPU memory is required #5

Closed amiltonwong closed 7 years ago

amiltonwong commented 7 years ago

Hi, authors,

How much GPU memory is required, in training and prediction, respectively? Does a 6GB GPU card meet the requirement?

THX!

GrassSunFlower commented 7 years ago

It depends on your patch size and way of usage. In terms of training, a 320320 patch with batch size 1 should take about 3GB. As for testing, a full resolution of 2048 1024 should take above 6GB GPU memory. So if you want to get better results, I would highly recommend you use a GPU with more than 8GB memory.