Newly implemented convnet (c++).

This is an EARLY ADOPTERS EDITION, which is still buggy, I'll post formal version in days.

To run this code, you should have

• a cifar-10 dataset;
• OpenCV.

##File description (todo)

##Compile & Run

• Compile: "cmake CMakeLists.txt" and then "make"

• Run: "./conv"

##Updates

• 3-channels images supported.
• Add Dropout;
• In conv layers, one can use either 3-channel conv kernels or single-chanel conv kernels (that is to say, whether share weights).
• Local Response Normalization supported (still buggy).