Detection good, split up col images

Joseph Redmon

2014-12-04 1edcf73a73d2007afc61289245763f5cf0c29e10

 src/network.c

@@ -31,146 +31,6 @@
    return net;
}

#ifdef GPU

void forward_network_gpu(network net, cl_mem input, cl_mem truth, int train)
{
    //printf("start\n");
    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);
            input = layer.output_cl;
        }
        else if(net.types[i] == COST){
            cost_layer layer = *(cost_layer *)net.layers[i];
            forward_cost_layer_gpu(layer, input, truth);
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            forward_connected_layer_gpu(layer, input);
            input = layer.output_cl;
        }
        else if(net.types[i] == MAXPOOL){
            maxpool_layer layer = *(maxpool_layer *)net.layers[i];
            forward_maxpool_layer_gpu(layer, input);
            input = layer.output_cl;
        }
        else if(net.types[i] == SOFTMAX){
            softmax_layer layer = *(softmax_layer *)net.layers[i];
            forward_softmax_layer_gpu(layer, input);
            input = layer.output_cl;
        }
        /*
           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] == NORMALIZATION){
           normalization_layer layer = *(normalization_layer *)net.layers[i];
           forward_normalization_layer(layer, input);
           input = layer.output;
           }
         */
    }
}

void backward_network_gpu(network net, cl_mem input)
{
    int i;
    cl_mem prev_input;
    cl_mem prev_delta;
    for(i = net.n-1; i >= 0; --i){
        if(i == 0){
            prev_input = input;
            prev_delta = 0;
        }else{
            prev_input = get_network_output_cl_layer(net, i-1);
            prev_delta = get_network_delta_cl_layer(net, i-1);
        }
        if(net.types[i] == CONVOLUTIONAL){
            convolutional_layer layer = *(convolutional_layer *)net.layers[i];
            backward_convolutional_layer_gpu(layer, prev_delta);
        }
        else if(net.types[i] == COST){
            cost_layer layer = *(cost_layer *)net.layers[i];
            backward_cost_layer_gpu(layer, prev_input, prev_delta);
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            backward_connected_layer_gpu(layer, prev_input, prev_delta);
        }
        else if(net.types[i] == MAXPOOL){
            maxpool_layer layer = *(maxpool_layer *)net.layers[i];
            backward_maxpool_layer_gpu(layer, prev_delta);
        }
        else if(net.types[i] == SOFTMAX){
            softmax_layer layer = *(softmax_layer *)net.layers[i];
            backward_softmax_layer_gpu(layer, prev_delta);
        }
    }
}

void update_network_gpu(network net)
{
    int i;
    for(i = 0; i < net.n; ++i){
        if(net.types[i] == CONVOLUTIONAL){
            convolutional_layer layer = *(convolutional_layer *)net.layers[i];
            update_convolutional_layer_gpu(layer);
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            update_connected_layer_gpu(layer);
        }
    }
}

cl_mem get_network_output_cl_layer(network net, int i)
{
    if(net.types[i] == CONVOLUTIONAL){
        convolutional_layer layer = *(convolutional_layer *)net.layers[i];
        return layer.output_cl;
    }
    else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.output_cl;
    }
    else if(net.types[i] == MAXPOOL){
        maxpool_layer layer = *(maxpool_layer *)net.layers[i];
        return layer.output_cl;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        return layer.output_cl;
    }
    return 0;
}

cl_mem get_network_delta_cl_layer(network net, int i)
{
    if(net.types[i] == CONVOLUTIONAL){
        convolutional_layer layer = *(convolutional_layer *)net.layers[i];
        return layer.delta_cl;
    }
    else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.delta_cl;
    }
    else if(net.types[i] == MAXPOOL){
        maxpool_layer layer = *(maxpool_layer *)net.layers[i];
        return layer.delta_cl;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        return layer.delta_cl;
    }
    return 0;
}

#endif

void forward_network(network net, float *input, float *truth, int train)
{
@@ -353,7 +213,7 @@
        }
        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];
@@ -379,43 +239,6 @@
}


#ifdef GPU
float train_network_datum_gpu(network net, float *x, float *y)
{
    int x_size = get_network_input_size(net)*net.batch;
    int y_size = get_network_output_size(net)*net.batch;
    if(!*net.input_cl){
        *net.input_cl = cl_make_array(x, x_size);
        *net.truth_cl = cl_make_array(y, y_size);
    }else{
        cl_write_array(*net.input_cl, x, x_size);
        cl_write_array(*net.truth_cl, y, y_size);
    }
    forward_network_gpu(net, *net.input_cl, *net.truth_cl, 1);
    backward_network_gpu(net, *net.input_cl);
    float error = get_network_cost(net);
    update_network_gpu(net);
    return error;
}

float train_network_sgd_gpu(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_batch(d, batch, X, y);
        float err = train_network_datum_gpu(net, X, y);
        sum += err;
    }
    free(X);
    free(y);
    return (float)sum/(n*batch);
}
#endif


float train_network_datum(network net, float *x, float *y)
@@ -438,7 +261,7 @@
    int i;
    float sum = 0;
    for(i = 0; i < n; ++i){
        get_batch(d, batch, X, y);
        get_random_batch(d, batch, X, y);
        float err = train_network_datum(net, X, y);
        sum += err;
    }
@@ -446,6 +269,7 @@
    free(y);
    return (float)sum/(n*batch);
}

float train_network_batch(network net, data d, int n)
{
    int i,j;
@@ -465,6 +289,23 @@
    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)
{
@@ -635,27 +476,14 @@
    } 
}

void top_predictions(network net, int n, int *index)
void top_predictions(network net, int k, int *index)
{
    int i,j;
    int k = get_network_output_size(net);
    int size = get_network_output_size(net);
    float *out = get_network_output(net);
    float thresh = FLT_MAX;
    for(i = 0; i < n; ++i){
        float max = -FLT_MAX;
        int max_i = -1;
        for(j = 0; j < k; ++j){
            float val = out[j];
            if(val > max &&  val < thresh){
                max = val;
                max_i = j;
            }
        }
        index[i] = max_i;
        thresh = max;
    }
    top_k(out, size, k, index);
}


float *network_predict(network net, float *input)
{
    forward_network(net, input, 0, 0);
@@ -693,7 +521,7 @@
    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;