direction of CNN

[Aaron723]

inspired by one hot encode, I think we can use multiple kernel at ONE layer to achieve better performance as a single kernel's learning ability is strongly limited

[Aaron723]

One possible solution: turn the output to four channels like 255 x 255 x 4. each pixel will be encoded into a one-hot vector(0,0,0,1), using 4 different kernels to do the convolution. Under the current solution, I found that one kernel can't learn sufficient features from the given input. Thus, it is import to enlarge the kernel size/increase the kernel number to solve this problem.