Incorrect Neural Network After New Run

Hi, I am a newbie with MATLAB but I have found things a bit weird in my process. I am training and testing a convolutional neural network approach. The flow is something like, train the network -> save the network struct in a .mat file, load the struct and run a test image on it.

Now this process would work well if they are done consecutively. But if I would preload the created network model and just run a new image on it, it would give some garbage results which is the same for all other new images.

The testing part only uses the actual network model and the image so I do not understand why this thing behaves as it is.

Anyone can explain this? Thank you!

[Greg Heath]

Insufficient details. Why not paste some code?