Running an ImageAI training returns always 0 loss

Question

I am trying the typical hololens tutorial to use Tensorflow, Keras, and ImageAI to be able to train a new custom object detection.

The training is running, but it is always returning something like:

loss: 0.0000e+00 - yolo_layer_1_loss: 0.0000e+00 - yolo_layer_2_loss: 0.0000e+00

And when I try to use the outcome of the model, it fails.

I am using Python 3.7.7 tensorflow 1.14.0
imageai 2.1.5
Keras 2.3.1

Answer 1

For me, what solved the issue was using batch size > 1.