So there WAS this huge bug. Gone now

Joseph Redmon

2014-05-09 cd8d53df21f3ad2810add2a8cff766c745f55a17

 src/network.c

@@ -6,18 +6,22 @@

#include "connected_layer.h"
#include "convolutional_layer.h"
//#include "old_conv.h"
#include "maxpool_layer.h"
#include "normalization_layer.h"
#include "softmax_layer.h"

network make_network(int n)
network make_network(int n, int batch)
{
    network net;
    net.n = n;
    net.batch = batch;
    net.layers = calloc(net.n, sizeof(void *));
    net.types = calloc(net.n, sizeof(LAYER_TYPE));
    net.outputs = 0;
    net.output = 0;
    #ifdef GPU
    net.input_cl = 0;
    #endif
    return net;
}

@@ -25,10 +29,11 @@
{
    int i;
    fprintf(fp, "[convolutional]\n");
    if(first) fprintf(fp,   "height=%d\n"
    if(first) fprintf(fp,   "batch=%d\n"
                            "height=%d\n"
                            "width=%d\n"
                            "channels=%d\n",
                            l->h, l->w, l->c);
                            l->batch,l->h, l->w, l->c);
    fprintf(fp, "filters=%d\n"
                "size=%d\n"
                "stride=%d\n"
@@ -44,11 +49,11 @@
{
    int i;
    fprintf(fp, "[connected]\n");
    if(first) fprintf(fp, "input=%d\n", l->inputs);
    if(first) fprintf(fp, "batch=%d\ninput=%d\n", l->batch, l->inputs);
    fprintf(fp, "output=%d\n"
                "activation=%s\n",
                l->outputs,
                get_activation_string(l->activation));
            "activation=%s\n",
            l->outputs,
            get_activation_string(l->activation));
    fprintf(fp, "data=");
    for(i = 0; i < l->outputs; ++i) fprintf(fp, "%g,", l->biases[i]);
    for(i = 0; i < l->inputs*l->outputs; ++i) fprintf(fp, "%g,", l->weights[i]);
@@ -58,17 +63,32 @@
void print_maxpool_cfg(FILE *fp, maxpool_layer *l, int first)
{
    fprintf(fp, "[maxpool]\n");
    if(first) fprintf(fp,   "height=%d\n"
                            "width=%d\n"
                            "channels=%d\n",
                            l->h, l->w, l->c);
    if(first) fprintf(fp,   "batch=%d\n"
            "height=%d\n"
            "width=%d\n"
            "channels=%d\n",
            l->batch,l->h, l->w, l->c);
    fprintf(fp, "stride=%d\n\n", l->stride);
}

void print_normalization_cfg(FILE *fp, normalization_layer *l, int first)
{
    fprintf(fp, "[localresponsenormalization]\n");
    if(first) fprintf(fp,   "batch=%d\n"
            "height=%d\n"
            "width=%d\n"
            "channels=%d\n",
            l->batch,l->h, l->w, l->c);
    fprintf(fp, "size=%d\n"
                "alpha=%g\n"
                "beta=%g\n"
                "kappa=%g\n\n", l->size, l->alpha, l->beta, l->kappa);
}

void print_softmax_cfg(FILE *fp, softmax_layer *l, int first)
{
    fprintf(fp, "[softmax]\n");
    if(first) fprintf(fp, "input=%d\n", l->inputs);
    if(first) fprintf(fp, "batch=%d\ninput=%d\n", l->batch, l->inputs);
    fprintf(fp, "\n");
}

@@ -85,24 +105,42 @@
            print_connected_cfg(fp, (connected_layer *)net.layers[i], i==0);
        else if(net.types[i] == MAXPOOL)
            print_maxpool_cfg(fp, (maxpool_layer *)net.layers[i], i==0);
        else if(net.types[i] == NORMALIZATION)
            print_normalization_cfg(fp, (normalization_layer *)net.layers[i], i==0);
        else if(net.types[i] == SOFTMAX)
            print_softmax_cfg(fp, (softmax_layer *)net.layers[i], i==0);
    }
    fclose(fp);
}

void forward_network(network net, float *input)
void forward_network(network net, float *input, int train)
{
    int i;
    #ifdef GPU
    cl_setup();
    size_t size = get_network_input_size(net);
    if(!net.input_cl){
        net.input_cl = clCreateBuffer(cl.context,
            CL_MEM_READ_WRITE, size*sizeof(float), 0, &cl.error);
        check_error(cl);
    }
    cl_write_array(net.input_cl, input, size);
    cl_mem input_cl = net.input_cl;
    #endif
    for(i = 0; i < net.n; ++i){
        if(net.types[i] == CONVOLUTIONAL){
            convolutional_layer layer = *(convolutional_layer *)net.layers[i];
            #ifdef GPU
            forward_convolutional_layer_gpu(layer, input_cl);
            input_cl = layer.output_cl;
            #else
            forward_convolutional_layer(layer, input);
            #endif
            input = layer.output;
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            forward_connected_layer(layer, input);
            forward_connected_layer(layer, input, train);
            input = layer.output;
        }
        else if(net.types[i] == SOFTMAX){
@@ -115,6 +153,11 @@
            forward_maxpool_layer(layer, input);
            input = layer.output;
        }
        else if(net.types[i] == NORMALIZATION){
            normalization_layer layer = *(normalization_layer *)net.layers[i];
            forward_normalization_layer(layer, input);
            input = layer.output;
        }
    }
}

@@ -132,6 +175,9 @@
        else if(net.types[i] == SOFTMAX){
            //maxpool_layer layer = *(maxpool_layer *)net.layers[i];
        }
        else if(net.types[i] == NORMALIZATION){
            //maxpool_layer layer = *(maxpool_layer *)net.layers[i];
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            update_connected_layer(layer, step, momentum, decay);
@@ -153,6 +199,9 @@
    } else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.output;
    } else if(net.types[i] == NORMALIZATION){
        normalization_layer layer = *(normalization_layer *)net.layers[i];
        return layer.output;
    }
    return 0;
}
@@ -191,11 +240,11 @@
    float *out = get_network_output(net);
    int i, k = get_network_output_size(net);
    for(i = 0; i < k; ++i){
        printf("%f, ", out[i]);
        //printf("%f, ", out[i]);
        delta[i] = truth[i] - out[i];
        sum += delta[i]*delta[i];
    }
    printf("\n");
    //printf("\n");
    return sum;
}

@@ -222,22 +271,23 @@
        }
        if(net.types[i] == CONVOLUTIONAL){
            convolutional_layer layer = *(convolutional_layer *)net.layers[i];
            learn_convolutional_layer(layer);
            //learn_convolutional_layer(layer);
            if(i != 0) backward_convolutional_layer(layer, prev_delta);
            backward_convolutional_layer(layer, prev_delta);
        }
        else if(net.types[i] == MAXPOOL){
            maxpool_layer layer = *(maxpool_layer *)net.layers[i];
            if(i != 0) backward_maxpool_layer(layer, prev_input, prev_delta);
        }
        else if(net.types[i] == NORMALIZATION){
            normalization_layer layer = *(normalization_layer *)net.layers[i];
            if(i != 0) backward_normalization_layer(layer, prev_input, prev_delta);
        }
        else if(net.types[i] == SOFTMAX){
            softmax_layer layer = *(softmax_layer *)net.layers[i];
            if(i != 0) backward_softmax_layer(layer, prev_input, prev_delta);
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            learn_connected_layer(layer, prev_input);
            if(i != 0) backward_connected_layer(layer, prev_input, prev_delta);
            backward_connected_layer(layer, prev_input, prev_delta);
        }
    }
    return error;
@@ -245,7 +295,7 @@

float train_network_datum(network net, float *x, float *y, float step, float momentum, float decay)
{
    forward_network(net, x);
    forward_network(net, x, 1);
    //int class = get_predicted_class_network(net);
    float error = backward_network(net, x, y);
    update_network(net, step, momentum, decay);
@@ -258,19 +308,26 @@
    int i;
    float error = 0;
    int correct = 0;
    int pos = 0;
    for(i = 0; i < n; ++i){
        int index = rand()%d.X.rows;
        error += train_network_datum(net, d.X.vals[index], d.y.vals[index], step, momentum, decay);
        float err = train_network_datum(net, d.X.vals[index], d.y.vals[index], step, momentum, decay);
        float *y = d.y.vals[index];
        int class = get_predicted_class_network(net);
        correct += (y[class]?1:0);
        if(y[1]){
            error += err;
            ++pos;
        }


        //printf("%d %f %f\n", i,net.output[0], d.y.vals[index][0]);
        //if((i+1)%10 == 0){
        //    printf("%d: %f\n", (i+1), (float)correct/(i+1));
        //}
    }
    printf("Accuracy: %f\n",(float) correct/n);
    return error/n;
    //printf("Accuracy: %f\n",(float) correct/n);
    return error/pos;
}
float train_network_batch(network net, data d, int n, float step, float momentum,float decay)
{
@@ -280,7 +337,7 @@
        int index = rand()%d.X.rows;
        float *x = d.X.vals[index];
        float *y = d.y.vals[index];
        forward_network(net, x);
        forward_network(net, x, 1);
        int class = get_predicted_class_network(net);
        backward_network(net, x, y);
        correct += (y[class]?1:0);
@@ -304,7 +361,28 @@
    }
    visualize_network(net);
    cvWaitKey(100);
    printf("Accuracy: %f\n", (float)correct/d.X.rows);
    fprintf(stderr, "Accuracy: %f\n", (float)correct/d.X.rows);
}

int get_network_input_size_layer(network net, int i)
{
    if(net.types[i] == CONVOLUTIONAL){
        convolutional_layer layer = *(convolutional_layer *)net.layers[i];
        return layer.h*layer.w*layer.c;
    }
    else if(net.types[i] == MAXPOOL){
        maxpool_layer layer = *(maxpool_layer *)net.layers[i];
        return layer.h*layer.w*layer.c;
    }
    else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        return layer.inputs;
    }
    return 0;
}

int get_network_output_size_layer(network net, int i)
@@ -330,29 +408,33 @@
    return 0;
}

int reset_network_size(network net, int h, int w, int c)
int resize_network(network net, int h, int w, int c)
{
    int i;
    for (i = 0; i < net.n; ++i){
        if(net.types[i] == CONVOLUTIONAL){
            convolutional_layer *layer = (convolutional_layer *)net.layers[i];
            layer->h = h;
            layer->w = w;
            layer->c = c;
            resize_convolutional_layer(layer, h, w, c);
            image output = get_convolutional_image(*layer);
            h = output.h;
            w = output.w;
            c = output.c;
        }
        else if(net.types[i] == MAXPOOL){
        }else if(net.types[i] == MAXPOOL){
            maxpool_layer *layer = (maxpool_layer *)net.layers[i];
            layer->h = h;
            layer->w = w;
            layer->c = c;
            resize_maxpool_layer(layer, h, w, c);
            image output = get_maxpool_image(*layer);
            h = output.h;
            w = output.w;
            c = output.c;
        }else if(net.types[i] == NORMALIZATION){
            normalization_layer *layer = (normalization_layer *)net.layers[i];
            resize_normalization_layer(layer, h, w, c);
            image output = get_normalization_image(*layer);
            h = output.h;
            w = output.w;
            c = output.c;
        }else{
            error("Cannot resize this type of layer");
        }
    }
    return 0;
@@ -364,6 +446,11 @@
    return get_network_output_size_layer(net, i);
}

int get_network_input_size(network net)
{
    return get_network_output_size_layer(net, 0);
}

image get_network_image_layer(network net, int i)
{
    if(net.types[i] == CONVOLUTIONAL){
@@ -374,6 +461,10 @@
        maxpool_layer layer = *(maxpool_layer *)net.layers[i];
        return get_maxpool_image(layer);
    }
    else if(net.types[i] == NORMALIZATION){
        normalization_layer layer = *(normalization_layer *)net.layers[i];
        return get_normalization_image(layer);
    }
    return make_empty_image(0,0,0);
}

@@ -389,20 +480,25 @@

void visualize_network(network net)
{
    image *prev = 0;
    int i;
    char buff[256];
    for(i = 0; i < net.n; ++i){
        sprintf(buff, "Layer %d", i);
        if(net.types[i] == CONVOLUTIONAL){
            convolutional_layer layer = *(convolutional_layer *)net.layers[i];
            visualize_convolutional_layer(layer, buff);
            prev = visualize_convolutional_layer(layer, buff, prev);
        }
        if(net.types[i] == NORMALIZATION){
            normalization_layer layer = *(normalization_layer *)net.layers[i];
            visualize_normalization_layer(layer, buff);
        }
    } 
}

float *network_predict(network net, float *input)
{
    forward_network(net, input);
    forward_network(net, input, 0);
    float *out = get_network_output(net);
    return out;
}
@@ -467,3 +563,4 @@
    return acc;
}