route handles input images well....ish

Joseph Redmon

2015-05-11 516f019ba6fb88de7218dd3b4eaeadb1cf676518

 src/network_kernels.cu

@@ -15,7 +15,6 @@
#include "deconvolutional_layer.h"
#include "maxpool_layer.h"
#include "cost_layer.h"
#include "normalization_layer.h"
#include "softmax_layer.h"
#include "dropout_layer.h"
#include "route_layer.h"
@@ -29,37 +28,29 @@
{
    int i;
    for(i = 0; i < net.n; ++i){
        if(net.types[i] == CONVOLUTIONAL){
            forward_convolutional_layer_gpu(*(convolutional_layer *)net.layers[i], state);
        layer l = net.layers[i];
        if(l.type == CONVOLUTIONAL){
            forward_convolutional_layer_gpu(l, state);
        } else if(l.type == DECONVOLUTIONAL){
            forward_deconvolutional_layer_gpu(l, state);
        } else if(l.type == DETECTION){
            forward_detection_layer_gpu(l, state);
        } else if(l.type == CONNECTED){
            forward_connected_layer_gpu(l, state);
        } else if(l.type == CROP){
            forward_crop_layer_gpu(l, state);
        } else if(l.type == COST){
            forward_cost_layer_gpu(l, state);
        } else if(l.type == SOFTMAX){
            forward_softmax_layer_gpu(l, state);
        } else if(l.type == MAXPOOL){
            forward_maxpool_layer_gpu(l, state);
        } else if(l.type == DROPOUT){
            forward_dropout_layer_gpu(l, state);
        } else if(l.type == ROUTE){
            forward_route_layer_gpu(l, net);
        }
        else if(net.types[i] == DECONVOLUTIONAL){
            forward_deconvolutional_layer_gpu(*(deconvolutional_layer *)net.layers[i], state);
        }
        else if(net.types[i] == COST){
            forward_cost_layer_gpu(*(cost_layer *)net.layers[i], state);
        }
        else if(net.types[i] == CONNECTED){
            forward_connected_layer_gpu(*(connected_layer *)net.layers[i], state);
        }
        else if(net.types[i] == DETECTION){
            forward_detection_layer_gpu(*(detection_layer *)net.layers[i], state);
        }
        else if(net.types[i] == MAXPOOL){
            forward_maxpool_layer_gpu(*(maxpool_layer *)net.layers[i], state);
        }
        else if(net.types[i] == SOFTMAX){
            forward_softmax_layer_gpu(*(softmax_layer *)net.layers[i], state);
        }
        else if(net.types[i] == DROPOUT){
            forward_dropout_layer_gpu(*(dropout_layer *)net.layers[i], state);
        }
        else if(net.types[i] == CROP){
            forward_crop_layer_gpu(*(crop_layer *)net.layers[i], state);
        }
        else if(net.types[i] == ROUTE){
            forward_route_layer_gpu(*(route_layer *)net.layers[i], net);
        }
        state.input = get_network_output_gpu_layer(net, i);
        state.input = l.output_gpu;
    }
}

@@ -68,40 +59,33 @@
    int i;
    float * original_input = state.input;
    for(i = net.n-1; i >= 0; --i){
        layer l = net.layers[i];
        if(i == 0){
            state.input = original_input;
            state.delta = 0;
        }else{
            state.input = get_network_output_gpu_layer(net, i-1);
            state.delta = get_network_delta_gpu_layer(net, i-1);
            layer prev = net.layers[i-1];
            state.input = prev.output_gpu;
            state.delta = prev.delta_gpu;
        }

        if(net.types[i] == CONVOLUTIONAL){
            backward_convolutional_layer_gpu(*(convolutional_layer *)net.layers[i], state);
        }
        else if(net.types[i] == DECONVOLUTIONAL){
            backward_deconvolutional_layer_gpu(*(deconvolutional_layer *)net.layers[i], state);
        }
        else if(net.types[i] == COST){
            backward_cost_layer_gpu(*(cost_layer *)net.layers[i], state);
        }
        else if(net.types[i] == CONNECTED){
            backward_connected_layer_gpu(*(connected_layer *)net.layers[i], state);
        }
        else if(net.types[i] == DETECTION){
            backward_detection_layer_gpu(*(detection_layer *)net.layers[i], state);
        }
        else if(net.types[i] == MAXPOOL){
            backward_maxpool_layer_gpu(*(maxpool_layer *)net.layers[i], state);
        }
        else if(net.types[i] == DROPOUT){
            backward_dropout_layer_gpu(*(dropout_layer *)net.layers[i], state);
        }
        else if(net.types[i] == SOFTMAX){
            backward_softmax_layer_gpu(*(softmax_layer *)net.layers[i], state);
        }
        else if(net.types[i] == ROUTE){
            backward_route_layer_gpu(*(route_layer *)net.layers[i], net);
        if(l.type == CONVOLUTIONAL){
            backward_convolutional_layer_gpu(l, state);
        } else if(l.type == DECONVOLUTIONAL){
            backward_deconvolutional_layer_gpu(l, state);
        } else if(l.type == MAXPOOL){
            if(i != 0) backward_maxpool_layer_gpu(l, state);
        } else if(l.type == DROPOUT){
            backward_dropout_layer_gpu(l, state);
        } else if(l.type == DETECTION){
            backward_detection_layer_gpu(l, state);
        } else if(l.type == SOFTMAX){
            if(i != 0) backward_softmax_layer_gpu(l, state);
        } else if(l.type == CONNECTED){
            backward_connected_layer_gpu(l, state);
        } else if(l.type == COST){
            backward_cost_layer_gpu(l, state);
        } else if(l.type == ROUTE){
            backward_route_layer_gpu(l, net);
        }
    }
}
@@ -111,89 +95,17 @@
    int i;
    int update_batch = net.batch*net.subdivisions;
    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, update_batch, net.learning_rate, net.momentum, net.decay);
        }
        else if(net.types[i] == DECONVOLUTIONAL){
            deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
            update_deconvolutional_layer_gpu(layer, net.learning_rate, net.momentum, net.decay);
        }
        else if(net.types[i] == CONNECTED){
            connected_layer layer = *(connected_layer *)net.layers[i];
            update_connected_layer_gpu(layer, update_batch, net.learning_rate, net.momentum, net.decay);
        layer l = net.layers[i];
        if(l.type == CONVOLUTIONAL){
            update_convolutional_layer_gpu(l, update_batch, net.learning_rate, net.momentum, net.decay);
        } else if(l.type == DECONVOLUTIONAL){
            update_deconvolutional_layer_gpu(l, net.learning_rate, net.momentum, net.decay);
        } else if(l.type == CONNECTED){
            update_connected_layer_gpu(l, update_batch, net.learning_rate, net.momentum, net.decay);
        }
    }
}

float * get_network_output_gpu_layer(network net, int i)
{
    if(net.types[i] == CONVOLUTIONAL){
        return ((convolutional_layer *)net.layers[i]) -> output_gpu;
    }
    else if(net.types[i] == DECONVOLUTIONAL){
        return ((deconvolutional_layer *)net.layers[i]) -> output_gpu;
    }
    else if(net.types[i] == DETECTION){
        return ((detection_layer *)net.layers[i]) -> output_gpu;
    }
    else if(net.types[i] == CONNECTED){
        return ((connected_layer *)net.layers[i]) -> output_gpu;
    }
    else if(net.types[i] == MAXPOOL){
        return ((maxpool_layer *)net.layers[i]) -> output_gpu;
    }
    else if(net.types[i] == CROP){
        return ((crop_layer *)net.layers[i]) -> output_gpu;
    }
    else if(net.types[i] == SOFTMAX){
        return ((softmax_layer *)net.layers[i]) -> output_gpu;
    }
    else if(net.types[i] == ROUTE){
        return ((route_layer *)net.layers[i]) -> output_gpu;
    }
    else if(net.types[i] == DROPOUT){
        return get_network_output_gpu_layer(net, i-1);
    }
    return 0;
}

float * get_network_delta_gpu_layer(network net, int i)
{
    if(net.types[i] == CONVOLUTIONAL){
        convolutional_layer layer = *(convolutional_layer *)net.layers[i];
        return layer.delta_gpu;
    }
    else if(net.types[i] == DETECTION){
        detection_layer layer = *(detection_layer *)net.layers[i];
        return layer.delta_gpu;
    }
    else if(net.types[i] == DECONVOLUTIONAL){
        deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
        return layer.delta_gpu;
    }
    else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.delta_gpu;
    }
    else if(net.types[i] == MAXPOOL){
        maxpool_layer layer = *(maxpool_layer *)net.layers[i];
        return layer.delta_gpu;
    }
    else if(net.types[i] == ROUTE){
        route_layer layer = *(route_layer *)net.layers[i];
        return layer.delta_gpu;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        return layer.delta_gpu;
    } else if(net.types[i] == DROPOUT){
        if(i == 0) return 0;
        return get_network_delta_gpu_layer(net, i-1);
    }
    return 0;
}

float train_network_datum_gpu(network net, float *x, float *y)
{
    network_state state;
@@ -219,33 +131,22 @@

float *get_network_output_layer_gpu(network net, int i)
{
    if(net.types[i] == CONVOLUTIONAL){
        convolutional_layer layer = *(convolutional_layer *)net.layers[i];
        return layer.output;
    }
    else if(net.types[i] == DECONVOLUTIONAL){
        deconvolutional_layer layer = *(deconvolutional_layer *)net.layers[i];
        return layer.output;
    }
    else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        cuda_pull_array(layer.output_gpu, layer.output, layer.outputs*layer.batch);
        return layer.output;
    }
    else if(net.types[i] == DETECTION){
        detection_layer layer = *(detection_layer *)net.layers[i];
        int outputs = get_detection_layer_output_size(layer);
        cuda_pull_array(layer.output_gpu, layer.output, outputs*layer.batch);
        return layer.output;
    }
    else if(net.types[i] == MAXPOOL){
        maxpool_layer layer = *(maxpool_layer *)net.layers[i];
        return layer.output;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        pull_softmax_layer_output(layer);
        return layer.output;
    layer l = net.layers[i];
    if(l.type == CONVOLUTIONAL){
        return l.output;
    } else if(l.type == DECONVOLUTIONAL){
        return l.output;
    } else if(l.type == CONNECTED){
        cuda_pull_array(l.output_gpu, l.output, l.outputs*l.batch);
        return l.output;
    } else if(l.type == DETECTION){
        cuda_pull_array(l.output_gpu, l.output, l.outputs*l.batch);
        return l.output;
    } else if(l.type == MAXPOOL){
        return l.output;
    } else if(l.type == SOFTMAX){
        pull_softmax_layer_output(l);
        return l.output;
    }
    return 0;
}
@@ -253,7 +154,7 @@
float *get_network_output_gpu(network net)
{
    int i;
    for(i = net.n-1; i > 0; --i) if(net.types[i] != COST) break;
    for(i = net.n-1; i > 0; --i) if(net.layers[i].type != COST) break;
    return get_network_output_layer_gpu(net, i);
}