void save_convolutional_weights_binary(layer l, FILE *fp)
{
#ifdef GPU
if(gpu_index >= 0){
pull_convolutional_layer(l);
}
#endif
binarize_filters(l.filters, l.n, l.c*l.size*l.size, l.binary_filters);
int size = l.c*l.size*l.size;
int i, j, k;
fwrite(l.biases, sizeof(float), l.n, fp);
if (l.batch_normalize){
fwrite(l.scales, sizeof(float), l.n, fp);
fwrite(l.rolling_mean, sizeof(float), l.n, fp);
fwrite(l.rolling_variance, sizeof(float), l.n, fp);
}
for(i = 0; i < l.n; ++i){
float mean = l.binary_filters[i*size];
if(mean < 0) mean = -mean;
fwrite(&mean, sizeof(float), 1, fp);
for(j = 0; j < size/8; ++j){
int index = i*size + j*8;
unsigned char c = 0;
for(k = 0; k < 8; ++k){
if (j*8 + k >= size) break;
if (l.binary_filters[index + k] > 0) c = (c | 1<<k);
}
fwrite(&c, sizeof(char), 1, fp);
}
}
}
void load_convolutional_weights(layer l, FILE *fp)
{
if(l.binary){
//load_convolutional_weights_binary(l, fp);
//return;
}
int num = l.n*l.c*l.size*l.size;
fread(l.biases, sizeof(float), l.n, fp);
if (l.batch_normalize && (!l.dontloadscales)){
fread(l.scales, sizeof(float), l.n, fp);
fread(l.rolling_mean, sizeof(float), l.n, fp);
fread(l.rolling_variance, sizeof(float), l.n, fp);
}
fread(l.filters, sizeof(float), num, fp);
if (l.flipped) {
transpose_matrix(l.filters, l.c*l.size*l.size, l.n);
}
if (l.binary) binarize_filters(l.filters, l.n, l.c*l.size*l.size, l.filters);
#ifdef GPU
if(gpu_index >= 0){
push_convolutional_layer(l);
}
#endif
}
void forward_convolutional_layer(convolutional_layer l, network_state state)
{
int out_h = convolutional_out_height(l);
int out_w = convolutional_out_width(l);
int i;
fill_cpu(l.outputs*l.batch, 0, l.output, 1);
/*
if(l.binary){
binarize_filters(l.filters, l.n, l.c*l.size*l.size, l.binary_filters);
binarize_filters2(l.filters, l.n, l.c*l.size*l.size, l.cfilters, l.scales);
swap_binary(&l);
}
*/
/*
if(l.binary){
int m = l.n;
int k = l.size*l.size*l.c;
int n = out_h*out_w;
char *a = l.cfilters;
float *b = state.workspace;
float *c = l.output;
for(i = 0; i < l.batch; ++i){
im2col_cpu(state.input, l.c, l.h, l.w,
l.size, l.stride, l.pad, b);
gemm_bin(m,n,k,1,a,k,b,n,c,n);
c += n*m;
state.input += l.c*l.h*l.w;
}
scale_bias(l.output, l.scales, l.batch, l.n, out_h*out_w);
add_bias(l.output, l.biases, l.batch, l.n, out_h*out_w);
activate_array(l.output, m*n*l.batch, l.activation);
return;
}
*/
if(l.xnor && (l.c%32 != 0 || !AI2)){
binarize_filters(l.filters, l.n, l.c*l.size*l.size, l.binary_filters);
swap_binary(&l);
for(i = 0; i < l.batch; ++i){
binarize_input(state.input + i*l.inputs, l.c, l.h*l.w, l.binary_input + i*l.inputs);
}
state.input = l.binary_input;
}
int m = l.n;
int k = l.size*l.size*l.c;
int n = out_h*out_w;
if (l.xnor && l.c%32 == 0 && AI2) {
forward_xnor_layer(l, state);
printf("xnor\n");
} else {
float *a = l.filters;
float *b = state.workspace;
float *c = l.output;
for(i = 0; i < l.batch; ++i){
im2col_cpu(state.input, l.c, l.h, l.w,
l.size, l.stride, l.pad, b);
gemm(0,0,m,n,k,1,a,k,b,n,1,c,n);
c += n*m;
state.input += l.c*l.h*l.w;
}
}
if(l.batch_normalize){
forward_batchnorm_layer(l, state);
}
add_bias(l.output, l.biases, l.batch, l.n, out_h*out_w);
activate_array(l.output, m*n*l.batch, l.activation);
if(l.binary || l.xnor) swap_binary(&l);
}
