tensorflow-cifar-10-NiN

Cifar-10 Network in Network implementation example using TensorFlow library.

Requirement

Library	Version
Python	^2.7
Tensorflow	^1.0.1
Numpy	^1.12.0
Pickle	*

Usage

Download code:

git clone https://github.com/eugenelet/tensorflow-cifar-10-NiN

cd tensorflow-cifar-10-NiN

Train cnn:

Batch size: 128

Prediction made on per epoch basis.

161 epochs takes about 3h on GTX 1080.

python train.py

Make prediction:

python predict.py

Example output:

Trying to restore last checkpoint ...
('Restored checkpoint from:', u'./tensorboard/aug-decay-RMS/-188692')
Accuracy on Test-Set: 86.18% (8618 / 10000)
[890   4  25   5  12   1  23   4  29   7] (0) airplane
[  8 895   2   3   3   3  30   2  18  36] (1) automobile
[ 19   0 759  15  59  24 106  13   4   1] (2) bird
[ 13   1  28 696  46  86  99  22   5   4] (3) cat
[  4   0  19  13 872  13  62  17   0   0] (4) deer
[  2   1  18  71  40 807  38  21   0   2] (5) dog
[  3   0   4   9   5   2 974   2   1   0] (6) frog
[  3   0  14  14  29  19  15 905   0   1] (7) horse
[ 32   1   4   3   8   3  26   0 913  10] (8) ship
[ 12  20   3   5   1   1  21   3  27 907] (9) truck
 (0) (1) (2) (3) (4) (5) (6) (7) (8) (9)

Tensorboard

tensorboard --logdir=./tensorboard

Model

Convolution layer 1
Conv_2d
ReLu
MLP
ReLu
MLP
ReLu
MaxPool
Convolution layer 2
Conv_2d
ReLu
MLP
ReLu
MLP
ReLu
MaxPool
Convolution layer 3
Conv_2d
ReLu
MLP
ReLu
MLP
ReLu
AvgPool
Softmax_linear

License

Apache License 2.0

Implementation Details

My Blog