ResNet 20/32/44/56/110 for CIFAR10 with caffe

1. get cifar10 python version, then create a soft link ln -s cifar-10-batches-py here
2. use data_utils.py to generate 4 pixel padded training data and testing data. Horizontal flip and random crop are performed on the fly while training.
3. use net_generator.py to generate solver.prototxt and trainval.prototxt, you can generate resnet or plain net of depth 20 44 56 110, or even deeper if you want. you just need to change n according to depth=6n+2
4. use train.sh to train it (eg., ./train.sh 0,1,2,3 resnet-20)
5. specify caffe path in cfgs.py and use plot.py to generate beautful loss plots.

results are consistent with original paper

seems there's no much difference between resnet-20 and plain-20. However, from the second plot, you can see that plain-110 have difficulty to converge.