1. Add download of data into ctc_read.py
2. Use 'clone' instead of 'get_inference_program'
3. Add time to train log

2, add examples for imagenet and mnist
3, support new operator: dropout, lrn, concat
4, compatible with old versions of caffe
5, fix bugs

格式修改

Reduce step time from 0.2 to 0.15

2. Add ceil mode for pooling