Pseudo labeling with multiple outputs

My question is for to the Fisheries Monitoring competition, but I think it general enough to get it’s own post.
How can I create pseudo labeled test_batches with multiple outputs (bounding box and class)?

I got the pseudo labeling predictions on the test set in preds
preds = model.predict(conv_test_feat, batch_size=batch_size*2)
preds[0] is the bounding box prediction and preds[1] is the class prediction

When I’m trying to get the test_batches by:
test_batches = gen.flow(conv_test_feat, preds, batch_size=16)

I receive the following error:
ValueError: could not broadcast input array from shape (1000,4) into shape (1000)