Hi,
Does Caffe support integer-based models?
If it does, I guess that:
- the weights in my .caffemodel must be integers
- the input data must be integer
- what are the supported integer formats (16bit, 32bit)?
- what is the format for the partial sum (for convolution you apply partial_sum += input * weight kernel_size times before activation function)?