I'm following example of DIGITS, object-detection

I’m following object detection example.

When I trained the network using pre-trained GoogleNet(.caffemodel) you linked, using epoch 10.

And I cloned the Job and trained Again.

I didn’t change any options, but the result is different.

The difference is not a little.

The first model’s mAP is 42, but the second model’ mAP is 0.

Why the difference is occurred.

I just clicked ‘clone job’ and ‘create’.