Development log

update .gitignore to ignore model and log data

utils/cifar_util.py

correct data size to original 50k
normalise data using calculated channel mean and standard deviation

utils/ops.py

add dropout operation
add scale parameter for batch normalisation

main.py

add automatic restoring from latest checkpoint
add a set of parameters for network.py to use
change experiment structure to
```
  python main.py --model_id xxx
```

The models will be saved in modeldir/xxx together with current main.py, network.py and ops.py.

network.py

add learning rate placeholder for scheduled learning rate
add L2 regularisation
add several layers to make the network standard VGG16-D
add dropout options
add one more fully connected layer
change last activation to softmax
add scheduled learning rate
Data augmentation
- add random flip
- add random crop
Testing
- add test all options, will test all checkpoints in of the given model
- add debug information if a single checkpoint is tested