From 336a19f14550ab1adbb0d9599284ac525fb6a5e0 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Thu, 08 Sep 2016 05:44:41 +0000
Subject: [PATCH] slight changes to default demo settings
---
src/convolutional_layer.c | 149 ++++++++++++++++++++++++-------------------------
1 files changed, 74 insertions(+), 75 deletions(-)
diff --git a/src/convolutional_layer.c b/src/convolutional_layer.c
index 4014a24..ad2d8a5 100644
--- a/src/convolutional_layer.c
+++ b/src/convolutional_layer.c
@@ -70,18 +70,12 @@
int convolutional_out_height(convolutional_layer l)
{
- int h = l.h;
- if (!l.pad) h -= l.size;
- else h -= 1;
- return h/l.stride + 1;
+ return (l.h + 2*l.pad - l.size) / l.stride + 1;
}
int convolutional_out_width(convolutional_layer l)
{
- int w = l.w;
- if (!l.pad) w -= l.size;
- else w -= 1;
- return w/l.stride + 1;
+ return (l.w + 2*l.pad - l.size) / l.stride + 1;
}
image get_convolutional_image(convolutional_layer l)
@@ -104,36 +98,37 @@
size_t get_workspace_size(layer l){
#ifdef CUDNN
- size_t most = 0;
- size_t s = 0;
- cudnnGetConvolutionForwardWorkspaceSize(cudnn_handle(),
- l.srcTensorDesc,
- l.filterDesc,
- l.convDesc,
- l.dstTensorDesc,
- l.fw_algo,
- &s);
- if (s > most) most = s;
- cudnnGetConvolutionBackwardFilterWorkspaceSize(cudnn_handle(),
- l.srcTensorDesc,
- l.ddstTensorDesc,
- l.convDesc,
- l.dfilterDesc,
- l.bf_algo,
- &s);
- if (s > most) most = s;
- cudnnGetConvolutionBackwardDataWorkspaceSize(cudnn_handle(),
- l.filterDesc,
- l.ddstTensorDesc,
- l.convDesc,
- l.dsrcTensorDesc,
- l.bd_algo,
- &s);
- if (s > most) most = s;
- return most;
-#else
+ if(gpu_index >= 0){
+ size_t most = 0;
+ size_t s = 0;
+ cudnnGetConvolutionForwardWorkspaceSize(cudnn_handle(),
+ l.srcTensorDesc,
+ l.filterDesc,
+ l.convDesc,
+ l.dstTensorDesc,
+ l.fw_algo,
+ &s);
+ if (s > most) most = s;
+ cudnnGetConvolutionBackwardFilterWorkspaceSize(cudnn_handle(),
+ l.srcTensorDesc,
+ l.ddstTensorDesc,
+ l.convDesc,
+ l.dfilterDesc,
+ l.bf_algo,
+ &s);
+ if (s > most) most = s;
+ cudnnGetConvolutionBackwardDataWorkspaceSize(cudnn_handle(),
+ l.filterDesc,
+ l.ddstTensorDesc,
+ l.convDesc,
+ l.dsrcTensorDesc,
+ l.bd_algo,
+ &s);
+ if (s > most) most = s;
+ return most;
+ }
+ #endif
return (size_t)l.out_h*l.out_w*l.size*l.size*l.c*sizeof(float);
-#endif
}
#ifdef GPU
@@ -147,8 +142,7 @@
cudnnSetTensor4dDescriptor(l->srcTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->c, l->h, l->w);
cudnnSetTensor4dDescriptor(l->dstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->out_c, l->out_h, l->out_w);
cudnnSetFilter4dDescriptor(l->filterDesc, CUDNN_DATA_FLOAT, CUDNN_TENSOR_NCHW, l->n, l->c, l->size, l->size);
- int padding = l->pad ? l->size/2 : 0;
- cudnnSetConvolution2dDescriptor(l->convDesc, padding, padding, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION);
+ cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION);
cudnnGetConvolutionForwardAlgorithm(cudnn_handle(),
l->srcTensorDesc,
l->filterDesc,
@@ -177,7 +171,7 @@
#endif
#endif
-convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int n, int size, int stride, int pad, ACTIVATION activation, int batch_normalize, int binary, int xnor)
+convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int n, int size, int stride, int padding, ACTIVATION activation, int batch_normalize, int binary, int xnor)
{
int i;
convolutional_layer l = {0};
@@ -192,7 +186,7 @@
l.batch = batch;
l.stride = stride;
l.size = size;
- l.pad = pad;
+ l.pad = padding;
l.batch_normalize = batch_normalize;
l.filters = calloc(c*n*size*size, sizeof(float));
@@ -240,49 +234,51 @@
}
#ifdef GPU
- l.filters_gpu = cuda_make_array(l.filters, c*n*size*size);
- l.filter_updates_gpu = cuda_make_array(l.filter_updates, c*n*size*size);
+ if(gpu_index >= 0){
+ l.filters_gpu = cuda_make_array(l.filters, c*n*size*size);
+ l.filter_updates_gpu = cuda_make_array(l.filter_updates, c*n*size*size);
- l.biases_gpu = cuda_make_array(l.biases, n);
- l.bias_updates_gpu = cuda_make_array(l.bias_updates, n);
+ l.biases_gpu = cuda_make_array(l.biases, n);
+ l.bias_updates_gpu = cuda_make_array(l.bias_updates, n);
- l.scales_gpu = cuda_make_array(l.scales, n);
- l.scale_updates_gpu = cuda_make_array(l.scale_updates, n);
+ l.scales_gpu = cuda_make_array(l.scales, n);
+ l.scale_updates_gpu = cuda_make_array(l.scale_updates, n);
- l.delta_gpu = cuda_make_array(l.delta, l.batch*out_h*out_w*n);
- l.output_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n);
+ l.delta_gpu = cuda_make_array(l.delta, l.batch*out_h*out_w*n);
+ l.output_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n);
- if(binary){
- l.binary_filters_gpu = cuda_make_array(l.filters, c*n*size*size);
- }
- if(xnor){
- l.binary_filters_gpu = cuda_make_array(l.filters, c*n*size*size);
- l.binary_input_gpu = cuda_make_array(0, l.inputs*l.batch);
- }
+ if(binary){
+ l.binary_filters_gpu = cuda_make_array(l.filters, c*n*size*size);
+ }
+ if(xnor){
+ l.binary_filters_gpu = cuda_make_array(l.filters, c*n*size*size);
+ l.binary_input_gpu = cuda_make_array(0, l.inputs*l.batch);
+ }
- if(batch_normalize){
- l.mean_gpu = cuda_make_array(l.mean, n);
- l.variance_gpu = cuda_make_array(l.variance, n);
+ if(batch_normalize){
+ l.mean_gpu = cuda_make_array(l.mean, n);
+ l.variance_gpu = cuda_make_array(l.variance, n);
- l.rolling_mean_gpu = cuda_make_array(l.mean, n);
- l.rolling_variance_gpu = cuda_make_array(l.variance, n);
+ l.rolling_mean_gpu = cuda_make_array(l.mean, n);
+ l.rolling_variance_gpu = cuda_make_array(l.variance, n);
- l.mean_delta_gpu = cuda_make_array(l.mean, n);
- l.variance_delta_gpu = cuda_make_array(l.variance, n);
+ l.mean_delta_gpu = cuda_make_array(l.mean, n);
+ l.variance_delta_gpu = cuda_make_array(l.variance, n);
- l.x_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n);
- l.x_norm_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n);
- }
+ l.x_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n);
+ l.x_norm_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n);
+ }
#ifdef CUDNN
- cudnnCreateTensorDescriptor(&l.srcTensorDesc);
- cudnnCreateTensorDescriptor(&l.dstTensorDesc);
- cudnnCreateFilterDescriptor(&l.filterDesc);
- cudnnCreateTensorDescriptor(&l.dsrcTensorDesc);
- cudnnCreateTensorDescriptor(&l.ddstTensorDesc);
- cudnnCreateFilterDescriptor(&l.dfilterDesc);
- cudnnCreateConvolutionDescriptor(&l.convDesc);
- cudnn_convolutional_setup(&l);
+ cudnnCreateTensorDescriptor(&l.srcTensorDesc);
+ cudnnCreateTensorDescriptor(&l.dstTensorDesc);
+ cudnnCreateFilterDescriptor(&l.filterDesc);
+ cudnnCreateTensorDescriptor(&l.dsrcTensorDesc);
+ cudnnCreateTensorDescriptor(&l.ddstTensorDesc);
+ cudnnCreateFilterDescriptor(&l.dfilterDesc);
+ cudnnCreateConvolutionDescriptor(&l.convDesc);
+ cudnn_convolutional_setup(&l);
#endif
+ }
#endif
l.workspace_size = get_workspace_size(l);
l.activation = activation;
@@ -301,6 +297,9 @@
l.filters[i*l.c*l.size*l.size + j] *= scale;
}
l.biases[i] -= l.rolling_mean[i] * scale;
+ l.scales[i] = 1;
+ l.rolling_mean[i] = 0;
+ l.rolling_variance[i] = 1;
}
}
@@ -434,7 +433,7 @@
}
*/
- if(l.xnor ){
+ if(l.xnor){
binarize_filters(l.filters, l.n, l.c*l.size*l.size, l.binary_filters);
swap_binary(&l);
binarize_cpu(state.input, l.c*l.h*l.w*l.batch, l.binary_input);
--
Gitblit v1.10.0