Closed XYAskWhy closed 6 years ago
If I don't get it wrong, you mean you want to train more than one image in a single GPU? It's possible to do so. See https://github.com/Zardinality/TF_Deformable_Net/issues/16. @Zardinality made a great answer for this problem.
@jessemelpolio I got it. Thank you.
Seemed that you can only accelerate training/testing by using multiple GPUs, which doesn't make sense now, since the memory of mainstream GPU is more than 10G, while training with batch_image 1 only use about 4G per card.