Attempt at visualizing ImageNet Features

Joseph Redmon

2014-04-11 cc06817efa24f20811ef6b32143c6700a91c5f2a

 src/convolutional_layer.c

@@ -285,52 +285,47 @@
    return float_to_image(h,w,c,layer.filters+i*h*w*c);
}

void visualize_convolutional_layer(convolutional_layer layer, char *window)
image *weighted_sum_filters(convolutional_layer layer, image *prev_filters)
{
    int color = 1;
    int border = 1;
    int h,w,c;
    int size = layer.size;
    h = size;
    w = (size + border) * layer.n - border;
    c = layer.c;
    if(c != 3 || !color){
        h = (h+border)*c - border;
        c = 1;
    image *filters = calloc(layer.n, sizeof(image));
    int i,j,k,c;
    if(!prev_filters){
        for(i = 0; i < layer.n; ++i){
            filters[i] = copy_image(get_convolutional_filter(layer, i));
        }
    }

    image filters = make_image(h,w,c);
    int i,j;
    for(i = 0; i < layer.n; ++i){
        int w_offset = i*(size+border);
        image k = get_convolutional_filter(layer, i);
        //printf("%f ** ", layer.biases[i]);
        //print_image(k);
        image copy = copy_image(k);
        normalize_image(copy);
        for(j = 0; j < k.c; ++j){
            //set_pixel(copy,0,0,j,layer.biases[i]);
        }
        if(c == 3 && color){
            embed_image(copy, filters, 0, w_offset);
        }
        else{
            for(j = 0; j < k.c; ++j){
                int h_offset = j*(size+border);
                image layer = get_image_layer(k, j);
                embed_image(layer, filters, h_offset, w_offset);
                free_image(layer);
    else{
        image base = prev_filters[0];
        for(i = 0; i < layer.n; ++i){
            image filter = get_convolutional_filter(layer, i);
            filters[i] = make_image(base.h, base.w, base.c);
            for(j = 0; j < layer.size; ++j){
                for(k = 0; k < layer.size; ++k){
                    for(c = 0; c < layer.c; ++c){
                        float weight = get_pixel(filter, j, k, c);
                        image prev_filter = copy_image(prev_filters[c]);
                        scale_image(prev_filter, weight);
                        add_into_image(prev_filter, filters[i], 0,0);
                        free_image(prev_filter);
                    }
                }
            }
        }
        free_image(copy);
    }
    return filters;
}

image *visualize_convolutional_layer(convolutional_layer layer, char *window, image *prev_filters)
{
    image *single_filters = weighted_sum_filters(layer, 0);
    show_images(single_filters, layer.n, window);

    image delta = get_convolutional_delta(layer);
    image dc = collapse_image_layers(delta, 1);
    char buff[256];
    sprintf(buff, "%s: Delta", window);
    show_image(dc, buff);
    //show_image(dc, buff);
    free_image(dc);
    show_image(filters, window);
    free_image(filters);
    return single_filters;
}