Multi-box loss dimensions about ssd_keras HOT 1 OPEN

@oarriaga Actually, these auxiliary zeros are needed for Keras. It cannot handle y_true and y_pred of different sizes, so, even if in y_true there is no need to provide priors coordinates, I have to preserve the shape. So, these zeros are not used, but are added for Keras compatability.

As for the second part of your question, probably, you're right. If we classify some prior as background prior, we can say, that is was not assigned as a ground truth prior. If you want to do this, you'll also need to change MultiboxLoss. But in any case, you need to output auxiliary zeros at last columns instead of priors for the reasons described above.

from ssd_keras.