Reproducing mnist results on new image data ?

[ahariri13]

Hi,
I am trying to reproduce the mnist results on images data of my own. They are simple parabolas (almost like the ones in the tutorial on jupyter notebook) and I am classifying them into 4 classes based on their curvature shape. I got good results using CNN with the same number of conv filters and FC layers as shown in the attached photo. Yet I am not sure on what to set for the polynomial orders in the parameters, as well as the coarsening level and number of edges in the arguments. I tried different values but my loss and accuracy remain the same in every epoch (attached as well). Any advice would be appreciated. thank you in advance !

[sousoufa]

Please can you help me
How i can use this work to classify my own images dataset