C1_W2_E2 Cannot get 99% accuracy

[XINYAN_ZHANG]

Hi all

Seems that my network cannot get 99% accuracy after even 10 epochs, really frustrated, I wonder why.

This is the output during training

Epoch 1/10
60000/60000 [==============================] - 10s 172us/sample - loss: 2.6560 - acc: 0.9071
Epoch 2/10
60000/60000 [==============================] - 9s 157us/sample - loss: 0.3395 - acc: 0.9366
Epoch 3/10
60000/60000 [==============================] - 10s 162us/sample - loss: 0.2845 - acc: 0.9428
Epoch 4/10
60000/60000 [==============================] - 10s 173us/sample - loss: 0.2716 - acc: 0.9444
Epoch 5/10
60000/60000 [==============================] - 10s 167us/sample - loss: 0.2324 - acc: 0.9520
Epoch 6/10
60000/60000 [==============================] - 10s 168us/sample - loss: 0.2328 - acc: 0.9535
Epoch 7/10
60000/60000 [==============================] - 10s 172us/sample - loss: 0.2196 - acc: 0.9563
Epoch 8/10
60000/60000 [==============================] - 10s 170us/sample - loss: 0.1922 - acc: 0.9593
Epoch 9/10
60000/60000 [==============================] - 10s 175us/sample - loss: 0.2053 - acc: 0.9599
Epoch 10/10
60000/60000 [==============================] - 10s 173us/sample - loss: 0.1937 - acc: 0.9613

Thanks

[XINYAN_ZHANG]

I have found the solution:

I need to add normalization (which is the gray scale value of images divided by 255.0) before feeding data into neural network.