yolo working w/ regions

Joseph Redmon

2015-09-16 c53e03348c65462bcba33f6352087dd6b9268e8f

 src/network.c

@@ -4,12 +4,14 @@
#include "image.h"
#include "data.h"
#include "utils.h"
#include "blas.h"

#include "crop_layer.h"
#include "connected_layer.h"
#include "convolutional_layer.h"
#include "deconvolutional_layer.h"
#include "detection_layer.h"
#include "region_layer.h"
#include "normalization_layer.h"
#include "maxpool_layer.h"
#include "avgpool_layer.h"
@@ -18,6 +20,41 @@
#include "dropout_layer.h"
#include "route_layer.h"

int get_current_batch(network net)
{
    int batch_num = (*net.seen)/(net.batch*net.subdivisions);
    return batch_num;
}

float get_current_rate(network net)
{
    int batch_num = get_current_batch(net);
    int i;
    float rate;
    switch (net.policy) {
        case CONSTANT:
            return net.learning_rate;
        case STEP:
            return net.learning_rate * pow(net.scale, batch_num/net.step);
        case STEPS:
            rate = net.learning_rate;
            for(i = 0; i < net.num_steps; ++i){
                if(net.steps[i] > batch_num) return rate;
                rate *= net.scales[i];
            }
            return rate;
        case EXP:
            return net.learning_rate * pow(net.gamma, batch_num);
        case POLY:
            return net.learning_rate * pow(1 - (float)batch_num / net.max_batches, net.power);
        case SIG:
            return net.learning_rate * (1./(1.+exp(net.gamma*(batch_num - net.step))));
        default:
            fprintf(stderr, "Policy is weird!\n");
            return net.learning_rate;
    }
}

char *get_layer_string(LAYER_TYPE a)
{
    switch(a){
@@ -35,6 +72,8 @@
            return "softmax";
        case DETECTION:
            return "detection";
        case REGION:
            return "region";
        case DROPOUT:
            return "dropout";
        case CROP:
@@ -56,6 +95,7 @@
    network net = {0};
    net.n = n;
    net.layers = calloc(net.n, sizeof(layer));
    net.seen = calloc(1, sizeof(int));
    #ifdef GPU
    net.input_gpu = calloc(1, sizeof(float *));
    net.truth_gpu = calloc(1, sizeof(float *));
@@ -79,6 +119,8 @@
            forward_normalization_layer(l, state);
        } else if(l.type == DETECTION){
            forward_detection_layer(l, state);
        } else if(l.type == REGION){
            forward_region_layer(l, state);
        } else if(l.type == CONNECTED){
            forward_connected_layer(l, state);
        } else if(l.type == CROP){
@@ -104,14 +146,15 @@
{
    int i;
    int update_batch = net.batch*net.subdivisions;
    float rate = get_current_rate(net);
    for(i = 0; i < net.n; ++i){
        layer l = net.layers[i];
        if(l.type == CONVOLUTIONAL){
            update_convolutional_layer(l, update_batch, net.learning_rate, net.momentum, net.decay);
            update_convolutional_layer(l, update_batch, rate, net.momentum, net.decay);
        } else if(l.type == DECONVOLUTIONAL){
            update_deconvolutional_layer(l, net.learning_rate, net.momentum, net.decay);
            update_deconvolutional_layer(l, rate, net.momentum, net.decay);
        } else if(l.type == CONNECTED){
            update_connected_layer(l, update_batch, net.learning_rate, net.momentum, net.decay);
            update_connected_layer(l, update_batch, rate, net.momentum, net.decay);
        }
    }
}
@@ -129,12 +172,16 @@
    float sum = 0;
    int count = 0;
    for(i = 0; i < net.n; ++i){
        if(net.layers[net.n-1].type == COST){
            sum += net.layers[net.n-1].output[0];
        if(net.layers[i].type == COST){
            sum += net.layers[i].output[0];
            ++count;
        }
        if(net.layers[net.n-1].type == DETECTION){
            sum += net.layers[net.n-1].cost[0];
        if(net.layers[i].type == DETECTION){
            sum += net.layers[i].cost[0];
            ++count;
        }
        if(net.layers[i].type == REGION){
            sum += net.layers[i].cost[0];
            ++count;
        }
    }
@@ -177,6 +224,8 @@
            backward_dropout_layer(l, state);
        } else if(l.type == DETECTION){
            backward_detection_layer(l, state);
        } else if(l.type == REGION){
            backward_region_layer(l, state);
        } else if(l.type == SOFTMAX){
            if(i != 0) backward_softmax_layer(l, state);
        } else if(l.type == CONNECTED){
@@ -191,6 +240,7 @@

float train_network_datum(network net, float *x, float *y)
{
    *net.seen += net.batch;
#ifdef GPU
    if(gpu_index >= 0) return train_network_datum_gpu(net, x, y);
#endif
@@ -202,7 +252,7 @@
    forward_network(net, state);
    backward_network(net, state);
    float error = get_network_cost(net);
    if((net.seen/net.batch)%net.subdivisions == 0) update_network(net);
    if(((*net.seen)/net.batch)%net.subdivisions == 0) update_network(net);
    return error;
}

@@ -215,7 +265,6 @@
    int i;
    float sum = 0;
    for(i = 0; i < n; ++i){
        net.seen += batch;
        get_random_batch(d, batch, X, y);
        float err = train_network_datum(net, X, y);
        sum += err;
@@ -236,7 +285,6 @@
    float sum = 0;
    for(i = 0; i < n; ++i){
        get_next_batch(d, batch, i*batch, X, y);
        net.seen += batch;
        float err = train_network_datum(net, X, y);
        sum += err;
    }
@@ -507,4 +555,17 @@
    return acc;
}


void free_network(network net)
{
    int i;
    for(i = 0; i < net.n; ++i){
        free_layer(net.layers[i]);
    }
    free(net.layers);
    #ifdef GPU
    if(*net.input_gpu) cuda_free(*net.input_gpu);
    if(*net.truth_gpu) cuda_free(*net.truth_gpu);
    if(net.input_gpu) free(net.input_gpu);
    if(net.truth_gpu) free(net.truth_gpu);
    #endif
}