lbls = lbls.to(device)
res = model(batch)
res = torch.sigmoid(res)
loss = wce(res, lbls, w_positive_tensor, w_negative_tensor)
optimizer.zero_grad()
loss.backward()
optimizer.step()
acc = torch.mean(
(torch.sum(lbls == (res > 0.5),
1) == predicted_size).type(torch.float32))
log.append_train(loss, acc)
model.eval()
N = len(testLoader)
for it, (batch, lbls) in enumerate(testLoader):
print(str(it) + '/' + str(N))
batch = batch.to(device)
lbls = lbls.to(device)
res = model(batch)
res = torch.sigmoid(res)
loss = wce(res, lbls, w_positive_tensor, w_negative_tensor)
acc = torch.mean(
