Open surajkumarpandey opened 2 years ago
It is trained on the MNIST data set as pointed out on the tutorial.
Thank you for clarifying. Could you also tell if the classification performance with 'real' camera images was the same as for the test dataset?
Hi, Thank you for the project. I was trying to use it with Rpi2. However, I am not able to get good classification results with my test samples. I wrote the digits on a paper and used it them as test data. The thresholding used in the project is not able to completely clean the image to give a 'MNIST like' image. There remain shadows and noise on the resultant image. If the model is trained on MNIST dataset and not actual camera images then I doubt the test and train spaces are not similar. Could you please tell if the training data used was the MNIST dataset or real camera images?