Use compute_loss_ota() if there is not loss_ota param or loss_ota==1

train.py

@@ -359,7 +359,7 @@ def train(hyp, opt, device, tb_writer=None):
             # Forward
             with amp.autocast(enabled=cuda):
                 pred = model(imgs)  # forward
-                if hyp['loss_ota'] == 1:
+                if 'loss_ota' not in hyp or hyp['loss_ota'] == 1:
                     loss, loss_items = compute_loss_ota(pred, targets.to(device), imgs)  # loss scaled by batch_size
                 else:
                     loss, loss_items = compute_loss(pred, targets.to(device))  # loss scaled by batch_size