25th place solution for Prostate cANcer graDe Assessment (PANDA) Challenge
Our code is based on Qishen Ha's code. Difference in our train code is
- batch_size : 2 => 8
- model : efficientnet-b0 => efficientnet-b1
We made a change in our inference code - TTA(Test Time Augmentation). You can see our inference code in below link.
We run our code with 4 V100 GPU. You can change GPU number in those two lines.
os.environ["CUDA_VISIBLE_DEVICES"] = "2,3,4,5"
model = nn.DataParallel(model, device_ids = [0,1,2,3])
Also, you can change fold number from 0 to 4 to select different fold as validation set.
fold = 0
At final submission, we use 2 weight file generated from fold=0 and fold=1.