if from sklearn.cross_validation import train_test_split gives error then use from sklearn.model_selection import train_test_split
A complete tutorial on using own dataset to train a CNN from scratch in Keras (TF & Theano Backend). Video Tutorial-part-1:https://www.youtube.com/edit?o=U&video_id=u8BW_fl6WRc
Video Tutorial-part-2:https://www.youtube.com/watch?v=NUuMg5m42-g&list=PLd9i_xMMzZF4fgKwm_OhCspy0DhvPbiHf&index=5
The updated_custom_data_cnn.py is explained in part 3 of the video which basically shows better way of assigning the labels to classes Video Tutorial-part-3:
Extract the data.zip and the python script in one directory.
if you would like to support conisder buying me a coffee