Visualizing with ipython, high accuracy on test but random prediction on ipython

[Hadi Keivan]

Hello,
I am facing a issue regarding visualization with ipython. The thing is that I am using a dataset generated using mnist dataset with Lenet architecture provided on caffe for a classification problem. However I obtained almost 90% of accuracy, but when I try to visualize it on ipython using and just modifying "classification" notebook example, the class predictions are so inaccurate. Any hints or ideas about what the issue might be would be really appreciated.
By the way, from the lenet architecture and deploy file, I just changed the source path in the data layer, kernel size and the number of output of the last fully connected layer.
Thnx

[Nam Vo]

Most likely bug on your python code, check the format of the input data.

[Hadi Keivan]

Thnx @Nam for your answer. however, I have generated different datasets RGB, Grayscale and in different sizes. I have converted them to "lmdb" Using both "create_mnist.sh" and "create_imagenet". I have also visualized the data and checked the labels and it seems that there is nothing wrong. Could you possibly have an idea of where exactly the issue could be so that I can recheck it?!
thanks again