"Size doesn't matter for me" - That's what our neural network said
Install them first on your laptop
scikit-images
tqdm
Took Mnist Data, resized it to 16x16, 24x24 and 28x28. Then I padded all of them to make it look like 32x32
Training Loss and Accuracy:
loss: 0.0620 - acc: 98.23
Validation Accuracy:
acc:98.44444444444445
To Run it just do:
python3 BaselineCNN.py
Testing:
X_tester = np.pad(resize(X_test[1174],(6,6)),[13,13],mode='constant')
np.argmax(model.predict(X_tester.reshape(1,32,32,1)))
plt.subplot(121)
plt.imshow(X_tester, cmap='gray')