Detection good, split up col images

Joseph Redmon

2014-12-04 1edcf73a73d2007afc61289245763f5cf0c29e10

 src/network.c

@@ -1,13 +1,17 @@
#include <stdio.h>
#include <time.h>
#include "network.h"
#include "image.h"
#include "data.h"
#include "utils.h"

#include "crop_layer.h"
#include "connected_layer.h"
#include "convolutional_layer.h"
#include "maxpool_layer.h"
#include "cost_layer.h"
#include "normalization_layer.h"
#include "freeweight_layer.h"
#include "softmax_layer.h"
#include "dropout_layer.h"

@@ -21,57 +25,14 @@
    net.outputs = 0;
    net.output = 0;
    #ifdef GPU
    net.input_cl = 0;
    net.input_cl = calloc(1, sizeof(cl_mem));
    net.truth_cl = calloc(1, sizeof(cl_mem));
    #endif
    return net;
}

#ifdef GPU
void forward_network(network net, float *input, int train)
{
    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;
    int i;
    for(i = 0; i < net.n; ++i){
        if(net.types[i] == CONVOLUTIONAL){
            convolutional_layer layer = *(convolutional_layer *)net.layers[i];
            forward_convolutional_layer_gpu(layer, input_cl);
            input_cl = layer.output_cl;
            input = layer.output;
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            forward_connected_layer(layer, input, train);
            input = layer.output;
        }
        else if(net.types[i] == SOFTMAX){
            softmax_layer layer = *(softmax_layer *)net.layers[i];
            forward_softmax_layer(layer, input);
            input = layer.output;
        }
        else if(net.types[i] == MAXPOOL){
            maxpool_layer layer = *(maxpool_layer *)net.layers[i];
            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;
        }
    }
}

#else

void forward_network(network net, float *input, int train)
void forward_network(network net, float *input, float *truth, int train)
{
    int i;
    for(i = 0; i < net.n; ++i){
@@ -85,6 +46,15 @@
            forward_connected_layer(layer, input);
            input = layer.output;
        }
        else if(net.types[i] == CROP){
            crop_layer layer = *(crop_layer *)net.layers[i];
            forward_crop_layer(layer, input);
            input = layer.output;
        }
        else if(net.types[i] == COST){
            cost_layer layer = *(cost_layer *)net.layers[i];
            forward_cost_layer(layer, input, truth);
        }
        else if(net.types[i] == SOFTMAX){
            softmax_layer layer = *(softmax_layer *)net.layers[i];
            forward_softmax_layer(layer, input);
@@ -105,9 +75,13 @@
            dropout_layer layer = *(dropout_layer *)net.layers[i];
            forward_dropout_layer(layer, input);
        }
        else if(net.types[i] == FREEWEIGHT){
            if(!train) continue;
            freeweight_layer layer = *(freeweight_layer *)net.layers[i];
            forward_freeweight_layer(layer, input);
        }
    }
}
#endif

void update_network(network net)
{
@@ -146,6 +120,8 @@
        return layer.output;
    } else if(net.types[i] == DROPOUT){
        return get_network_output_layer(net, i-1);
    } else if(net.types[i] == FREEWEIGHT){
        return get_network_output_layer(net, i-1);
    } else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.output;
@@ -157,7 +133,9 @@
}
float *get_network_output(network net)
{
    return get_network_output_layer(net, net.n-1);
    int i;
    for(i = net.n-1; i > 0; --i) if(net.types[i] != COST) break;
    return get_network_output_layer(net, i);
}

float *get_network_delta_layer(network net, int i)
@@ -173,6 +151,8 @@
        return layer.delta;
    } else if(net.types[i] == DROPOUT){
        return get_network_delta_layer(net, i-1);
    } else if(net.types[i] == FREEWEIGHT){
        return get_network_delta_layer(net, i-1);
    } else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.delta;
@@ -180,6 +160,14 @@
    return 0;
}

float get_network_cost(network net)
{
    if(net.types[net.n-1] == COST){
        return ((cost_layer *)net.layers[net.n-1])->output[0];
    }
    return 0;
}

float *get_network_delta(network net)
{
    return get_network_delta_layer(net, net.n-1);
@@ -210,9 +198,8 @@
    return max_index(out, k);
}

float backward_network(network net, float *input, float *truth)
void backward_network(network net, float *input)
{
    float error = calculate_error_network(net, truth);
    int i;
    float *prev_input;
    float *prev_delta;
@@ -226,11 +213,11 @@
        }
        if(net.types[i] == CONVOLUTIONAL){
            convolutional_layer layer = *(convolutional_layer *)net.layers[i];
            backward_convolutional_layer(layer, prev_delta);
            backward_convolutional_layer(layer, prev_input, 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);
            if(i != 0) backward_maxpool_layer(layer, prev_delta);
        }
        else if(net.types[i] == NORMALIZATION){
            normalization_layer layer = *(normalization_layer *)net.layers[i];
@@ -238,21 +225,28 @@
        }
        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);
            if(i != 0) backward_softmax_layer(layer, prev_delta);
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            backward_connected_layer(layer, prev_input, prev_delta);
        }
        else if(net.types[i] == COST){
            cost_layer layer = *(cost_layer *)net.layers[i];
            backward_cost_layer(layer, prev_input, prev_delta);
        }
    }
    return error;
}




float train_network_datum(network net, float *x, float *y)
{
    forward_network(net, x, 1);
    forward_network(net, x, y, 1);
    //int class = get_predicted_class_network(net);
    float error = backward_network(net, x, y);
    backward_network(net, x);
    float error = get_network_cost(net);
    update_network(net);
    //return (y[class]?1:0);
    return error;
@@ -264,46 +258,18 @@
    float *X = calloc(batch*d.X.cols, sizeof(float));
    float *y = calloc(batch*d.y.cols, sizeof(float));

    int i,j;
    int i;
    float sum = 0;
    for(i = 0; i < n; ++i){
        for(j = 0; j < batch; ++j){
            int index = rand()%d.X.rows;
            memcpy(X+j*d.X.cols, d.X.vals[index], d.X.cols*sizeof(float));
            memcpy(y+j*d.y.cols, d.y.vals[index], d.y.cols*sizeof(float));
        }
        get_random_batch(d, batch, X, y);
        float err = train_network_datum(net, X, y);
        sum += err;
        //train_network_datum(net, X, y);
        /*
        float *y = d.y.vals[index];
        int class = get_predicted_class_network(net);
        correct += (y[class]?1:0);
        */

/*
        for(j = 0; j < d.y.cols*batch; ++j){
            printf("%6.3f ", y[j]);
        }
        printf("\n");
        for(j = 0; j < d.y.cols*batch; ++j){
            printf("%6.3f ", get_network_output(net)[j]);
        }
        printf("\n");
        printf("\n");
        */


        //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);
    free(X);
    free(y);
    return (float)sum/(n*batch);
}

float train_network_batch(network net, data d, int n)
{
    int i,j;
@@ -314,14 +280,32 @@
            int index = rand()%d.X.rows;
            float *x = d.X.vals[index];
            float *y = d.y.vals[index];
            forward_network(net, x, 1);
            sum += backward_network(net, x, y);
            forward_network(net, x, y, 1);
            backward_network(net, x);
            sum += get_network_cost(net);
        }
        update_network(net);
    }
    return (float)sum/(n*batch);
}

float train_network_data_cpu(network net, data d, int n)
{
    int batch = net.batch;
    float *X = calloc(batch*d.X.cols, sizeof(float));
    float *y = calloc(batch*d.y.cols, sizeof(float));

    int i;
    float sum = 0;
    for(i = 0; i < n; ++i){
        get_next_batch(d, batch, i*batch, X, y);
        float err = train_network_datum(net, X, y);
        sum += err;
    }
    free(X);
    free(y);
    return (float)sum/(n*batch);
}

void train_network(network net, data d)
{
@@ -356,6 +340,10 @@
        dropout_layer layer = *(dropout_layer *) net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == FREEWEIGHT){
        freeweight_layer layer = *(freeweight_layer *) net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        return layer.inputs;
@@ -378,10 +366,15 @@
    else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.outputs;
    } else if(net.types[i] == DROPOUT){
    }
    else if(net.types[i] == DROPOUT){
        dropout_layer layer = *(dropout_layer *) net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == FREEWEIGHT){
        freeweight_layer layer = *(freeweight_layer *) net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        return layer.inputs;
@@ -423,7 +416,8 @@

int get_network_output_size(network net)
{
    int i = net.n-1;
    int i;
    for(i = net.n-1; i > 0; --i) if(net.types[i] != COST) break;
    return get_network_output_size_layer(net, i);
}

@@ -446,6 +440,10 @@
        normalization_layer layer = *(normalization_layer *)net.layers[i];
        return get_normalization_image(layer);
    }
    else if(net.types[i] == CROP){
        crop_layer layer = *(crop_layer *)net.layers[i];
        return get_crop_image(layer);
    }
    return make_empty_image(0,0,0);
}

@@ -464,6 +462,7 @@
    image *prev = 0;
    int i;
    char buff[256];
    //show_image(get_network_image_layer(net, 0), "Crop");
    for(i = 0; i < net.n; ++i){
        sprintf(buff, "Layer %d", i);
        if(net.types[i] == CONVOLUTIONAL){
@@ -477,19 +476,52 @@
    } 
}

void top_predictions(network net, int k, int *index)
{
    int size = get_network_output_size(net);
    float *out = get_network_output(net);
    top_k(out, size, k, index);
}


float *network_predict(network net, float *input)
{
    forward_network(net, input, 0);
    forward_network(net, input, 0, 0);
    float *out = get_network_output(net);
    return out;
}

matrix network_predict_data_multi(network net, data test, int n)
{
    int i,j,b,m;
    int k = get_network_output_size(net);
    matrix pred = make_matrix(test.X.rows, k);
    float *X = calloc(net.batch*test.X.rows, sizeof(float));
    for(i = 0; i < test.X.rows; i += net.batch){
        for(b = 0; b < net.batch; ++b){
            if(i+b == test.X.rows) break;
            memcpy(X+b*test.X.cols, test.X.vals[i+b], test.X.cols*sizeof(float));
        }
        for(m = 0; m < n; ++m){
            float *out = network_predict(net, X);
            for(b = 0; b < net.batch; ++b){
                if(i+b == test.X.rows) break;
                for(j = 0; j < k; ++j){
                    pred.vals[i+b][j] += out[j+b*k]/n;
                }
            }
        }
    }
    free(X);
    return pred;   
}

matrix network_predict_data(network net, data test)
{
    int i,j,b;
    int k = get_network_output_size(net);
    matrix pred = make_matrix(test.X.rows, k);
    float *X = calloc(net.batch*test.X.rows, sizeof(float));
    float *X = calloc(net.batch*test.X.cols, sizeof(float));
    for(i = 0; i < test.X.rows; i += net.batch){
        for(b = 0; b < net.batch; ++b){
            if(i+b == test.X.rows) break;
@@ -525,6 +557,12 @@
            image m = get_maxpool_image(layer);
            n = m.h*m.w*m.c;
        }
        else if(net.types[i] == CROP){
            crop_layer layer = *(crop_layer *)net.layers[i];
            output = layer.output;
            image m = get_crop_image(layer);
            n = m.h*m.w*m.c;
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            output = layer.output;
@@ -553,4 +591,12 @@
    return acc;
}

float network_accuracy_multi(network net, data d, int n)
{
    matrix guess = network_predict_data_multi(net, d, n);
    float acc = matrix_accuracy(d.y, guess);
    free_matrix(guess);
    return acc;
}