mirror/darknet
1
0
Fork 0
mirror of https://github.com/pjreddie/darknet.git synced 2023-08-10 21:13:14 +03:00
Code Issues Projects Releases Wiki Activity

Faster batch normalization

Browse Source
This commit is contained in:
Joseph Redmon
2015-12-07 17:18:04 -08:00
parent b4d81aa7bd
commit c2738835f0
7 changed files with 149 additions and 100 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/convolutional_kernels.cu
View File

@ -119,17 +119,14 @@ void forward_convolutional_layer_gpu(convolutional_layer l, network_state state)
if(l.batch_normalize){
if(state.train){
fast_mean_gpu(l.output_gpu, l.batch, l.n, l.out_h*l.out_w, l.spatial_mean_gpu, l.mean_gpu);
fast_variance_gpu(l.output_gpu, l.mean_gpu, l.batch, l.n, l.out_h*l.out_w, l.spatial_variance_gpu, l.variance_gpu);
fast_mean_gpu(l.output_gpu, l.batch, l.n, l.out_h*l.out_w, l.mean_gpu);
fast_variance_gpu(l.output_gpu, l.mean_gpu, l.batch, l.n, l.out_h*l.out_w, l.variance_gpu);
scal_ongpu(l.n, .95, l.rolling_mean_gpu, 1);
axpy_ongpu(l.n, .05, l.mean_gpu, 1, l.rolling_mean_gpu, 1);
scal_ongpu(l.n, .95, l.rolling_variance_gpu, 1);
axpy_ongpu(l.n, .05, l.variance_gpu, 1, l.rolling_variance_gpu, 1);
// cuda_pull_array(l.variance_gpu, l.mean, l.n);
// printf("%f\n", l.mean[0]);
copy_ongpu(l.outputs*l.batch, l.output_gpu, 1, l.x_gpu, 1);
normalize_gpu(l.output_gpu, l.mean_gpu, l.variance_gpu, l.batch, l.n, l.out_h*l.out_w);
copy_ongpu(l.outputs*l.batch, l.output_gpu, 1, l.x_norm_gpu, 1);
@ -161,8 +158,8 @@ void backward_convolutional_layer_gpu(convolutional_layer l, network_state state
scale_bias_gpu(l.delta_gpu, l.scales_gpu, l.batch, l.n, l.out_h*l.out_w);
fast_mean_delta_gpu(l.delta_gpu, l.variance_gpu, l.batch, l.n, l.out_w*l.out_h, l.spatial_mean_delta_gpu, l.mean_delta_gpu);
fast_variance_delta_gpu(l.x_gpu, l.delta_gpu, l.mean_gpu, l.variance_gpu, l.batch, l.n, l.out_w*l.out_h, l.spatial_variance_delta_gpu, l.variance_delta_gpu);
fast_mean_delta_gpu(l.delta_gpu, l.variance_gpu, l.batch, l.n, l.out_w*l.out_h, l.mean_delta_gpu);
fast_variance_delta_gpu(l.x_gpu, l.delta_gpu, l.mean_gpu, l.variance_gpu, l.batch, l.n, l.out_w*l.out_h, l.variance_delta_gpu);
normalize_delta_gpu(l.x_gpu, l.mean_gpu, l.variance_gpu, l.mean_delta_gpu, l.variance_delta_gpu, l.batch, l.n, l.out_w*l.out_h, l.delta_gpu);
}