Faster batch normalization

src/blas.h

@@ -36,14 +36,12 @@ void mean_gpu(float *x, int batch, int filters, int spatial, float *mean);
 void variance_gpu(float *x, float *mean, int batch, int filters, int spatial, float *variance);
 void normalize_gpu(float *x, float *mean, float *variance, int batch, int filters, int spatial);
 
-void mean_delta_gpu(float *delta, float *variance, int batch, int filters, int spatial, float *mean_delta);
-void variance_delta_gpu(float *x, float *delta, float *mean, float *variance, int batch, int filters, int spatial, float *variance_delta);
 void normalize_delta_gpu(float *x, float *mean, float *variance, float *mean_delta, float *variance_delta, int batch, int filters, int spatial, float *delta);
 
-void fast_mean_delta_gpu(float *delta, float *variance, int batch, int filters, int spatial, float *spatial_mean_delta, float *mean_delta);
-void fast_variance_delta_gpu(float *x, float *delta, float *mean, float *variance, int batch, int filters, int spatial, float *spatial_variance_delta, float *variance_delta);
+void fast_mean_delta_gpu(float *delta, float *variance, int batch, int filters, int spatial, float *mean_delta);
+void fast_variance_delta_gpu(float *x, float *delta, float *mean, float *variance, int batch, int filters, int spatial, float *variance_delta);
 
-void fast_variance_gpu(float *x, float *mean, int batch, int filters, int spatial, float *spatial_variance, float *variance);
-void fast_mean_gpu(float *x, int batch, int filters, int spatial, float *spatial_mean, float *mean);
+void fast_variance_gpu(float *x, float *mean, int batch, int filters, int spatial, float *variance);
+void fast_mean_gpu(float *x, int batch, int filters, int spatial, float *mean);
 #endif
 #endif

src/blas_kernels.cu

@@ -48,28 +48,6 @@ __global__ void  variance_delta_kernel(float *x, float *delta, float *mean, floa
     variance_delta[i] *= -.5 * pow(variance[i] + .00001f, (float)(-3./2.));
 }
 
-__global__ void spatial_variance_delta_kernel(float *x, float *delta, float *mean, float *variance, int batch, int filters, int spatial, float *spatial_variance_delta)
-{
-    int i = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
-    if (i >= batch*filters) return;
-    int f = i%filters;
-    int b = i/filters;
-
-    int k;
-    spatial_variance_delta[i] = 0;
-    for (k = 0; k < spatial; ++k) {
-        int index = b*filters*spatial + f*spatial + k;
-        spatial_variance_delta[i] += delta[index]*(x[index] - mean[f]);
-    }
-    spatial_variance_delta[i] *= -.5 * pow(variance[f] + .00001f, (float)(-3./2.));
-}
-
-extern "C" void variance_delta_gpu(float *x, float *delta, float *mean, float *variance, int batch, int filters, int spatial, float *variance_delta)
-{
-    variance_delta_kernel<<<cuda_gridsize(filters), BLOCK>>>(x, delta, mean, variance, batch, filters, spatial, variance_delta);
-    check_error(cudaPeekAtLastError());
-}
-
 __global__ void accumulate_kernel(float *x, int n, int groups, float *sum)
 {
     int k;
@@ -81,38 +59,62 @@ __global__ void accumulate_kernel(float *x, int n, int groups, float *sum)
     }
 }
 
-extern "C" void fast_variance_delta_gpu(float *x, float *delta, float *mean, float *variance, int batch, int filters, int spatial, float *spatial_variance_delta, float *variance_delta)
+__global__ void fast_mean_delta_kernel(float *delta, float *variance, int batch, int filters, int spatial, float *mean_delta)
 {
-    spatial_variance_delta_kernel<<<cuda_gridsize(filters*batch), BLOCK>>>(x, delta, mean, variance, batch, filters, spatial, spatial_variance_delta);
-    check_error(cudaPeekAtLastError());
-    accumulate_kernel<<<cuda_gridsize(filters), BLOCK>>>(spatial_variance_delta, batch, filters, variance_delta);
-    check_error(cudaPeekAtLastError());
-}
+    const int threads = BLOCK;
+    __shared__ float local[threads];
 
-__global__ void spatial_mean_delta_kernel(float *delta, float *variance, int batch, int filters, int spatial, float *spatial_mean_delta)
-{
-    int i = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
-    if (i >= batch*filters) return;
-    int f = i%filters;
-    int b = i/filters;
+    int id = threadIdx.x;
+    local[id] = 0;
 
-    int k;
-    spatial_mean_delta[i] = 0;
-    for (k = 0; k < spatial; ++k) {
-        int index = b*filters*spatial + f*spatial + k;
-        spatial_mean_delta[i] += delta[index];
+    int filter = blockIdx.x;
+
+    int i, j;
+    for(j = 0; j < batch; ++j){
+        for(i = 0; i < spatial; i += threads){
+            int index = j*spatial*filters + filter*spatial + i + id;
+            local[id] += (i+id < spatial) ? delta[index] : 0;
+        }
+    }
+
+    if(id == 0){
+        mean_delta[filter] = 0;
+        for(i = 0; i < threads; ++i){
+            mean_delta[filter] += local[i];
+        }
+        mean_delta[filter] *= (-1./sqrt(variance[filter] + .00001f));
     }
-    spatial_mean_delta[i] *= (-1./sqrt(variance[f] + .00001f));
 }
 
-extern "C" void fast_mean_delta_gpu(float *delta, float *variance, int batch, int filters, int spatial, float *spatial_mean_delta, float *mean_delta)
+__global__ void  fast_variance_delta_kernel(float *x, float *delta, float *mean, float *variance, int batch, int filters, int spatial, float *variance_delta)
 {
-    spatial_mean_delta_kernel<<<cuda_gridsize(filters*batch), BLOCK>>>(delta, variance, batch, filters, spatial, spatial_mean_delta);
-    check_error(cudaPeekAtLastError());
-    accumulate_kernel<<<cuda_gridsize(filters), BLOCK>>>(spatial_mean_delta, batch, filters, mean_delta);
-    check_error(cudaPeekAtLastError());
+    const int threads = BLOCK;
+    __shared__ float local[threads];
+
+    int id = threadIdx.x;
+    local[id] = 0;
+
+    int filter = blockIdx.x;
+
+    int i, j;
+    for(j = 0; j < batch; ++j){
+        for(i = 0; i < spatial; i += threads){
+            int index = j*spatial*filters + filter*spatial + i + id;
+
+            local[id] += (i+id < spatial) ? delta[index]*(x[index] - mean[filter]) : 0;
+        }
+    }
+
+    if(id == 0){
+        variance_delta[filter] = 0;
+        for(i = 0; i < threads; ++i){
+            variance_delta[filter] += local[i];
+        }
+        variance_delta[filter] *= -.5 * pow(variance[filter] + .00001f, (float)(-3./2.));
+    }
 }
 
+
 __global__ void mean_delta_kernel(float *delta, float *variance, int batch, int filters, int spatial, float *mean_delta)
 {
     int i = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
@@ -134,6 +136,18 @@ extern "C" void mean_delta_gpu(float *delta, float *variance, int batch, int fil
     check_error(cudaPeekAtLastError());
 }
 
+extern "C" void fast_mean_delta_gpu(float *delta, float *variance, int batch, int filters, int spatial, float *mean_delta)
+{
+    fast_mean_delta_kernel<<<filters, BLOCK>>>(delta, variance, batch, filters, spatial, mean_delta);
+    check_error(cudaPeekAtLastError());
+}
+
+extern "C" void fast_variance_delta_gpu(float *x, float *delta, float *mean, float *variance, int batch, int filters, int spatial, float *variance_delta)
+{
+    fast_variance_delta_kernel<<<filters, BLOCK>>>(x, delta, mean, variance, batch, filters, spatial, variance_delta);
+    check_error(cudaPeekAtLastError());
+}
+
 __global__ void  mean_kernel(float *x, int batch, int filters, int spatial, float *mean)
 {
     float scale = 1./(batch * spatial);
@@ -150,23 +164,6 @@ __global__ void  mean_kernel(float *x, int batch, int filters, int spatial, floa
     mean[i] *= scale;
 }
 
-__global__ void spatial_variance_kernel(float *x, float *mean, int batch, int filters, int spatial, float *variance)
-{
-    float scale = 1./(spatial*batch-1);
-    int k;
-    int i = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
-    if (i >= batch*filters) return;
-    int f = i%filters;
-    int b = i/filters;
-
-    variance[i] = 0;
-    for(k = 0; k < spatial; ++k){
-        int index = b*filters*spatial + f*spatial + k;
-        variance[i] += pow((x[index] - mean[f]), 2);
-    }
-    variance[i] *= scale;
-}
-
 __global__ void variance_kernel(float *x, float *mean, int batch, int filters, int spatial, float *variance)
 {
     float scale = 1./(batch * spatial);
@@ -238,28 +235,80 @@ extern "C" void normalize_gpu(float *x, float *mean, float *variance, int batch,
     check_error(cudaPeekAtLastError());
 }
 
+__global__ void  fast_mean_kernel(float *x, int batch, int filters, int spatial, float *mean)
+{
+    const int threads = BLOCK;
+    __shared__ float local[threads];
+
+    int id = threadIdx.x;
+    local[id] = 0;
+
+    int filter = blockIdx.x;
+
+    int i, j;
+    for(j = 0; j < batch; ++j){
+        for(i = 0; i < spatial; i += threads){
+            int index = j*spatial*filters + filter*spatial + i + id;
+            local[id] += (i+id < spatial) ? x[index] : 0;
+        }
+    }
+
+    if(id == 0){
+        mean[filter] = 0;
+        for(i = 0; i < threads; ++i){
+            mean[filter] += local[i];
+        }
+        mean[filter] /= spatial * batch;
+    }
+}
+
+__global__ void  fast_variance_kernel(float *x, float *mean, int batch, int filters, int spatial, float *variance)
+{
+    const int threads = BLOCK;
+    __shared__ float local[threads];
+
+    int id = threadIdx.x;
+    local[id] = 0;
+
+    int filter = blockIdx.x;
+
+    int i, j;
+    for(j = 0; j < batch; ++j){
+        for(i = 0; i < spatial; i += threads){
+            int index = j*spatial*filters + filter*spatial + i + id;
+
+            local[id] += (i+id < spatial) ? pow((x[index] - mean[filter]), 2) : 0;
+        }
+    }
+
+    if(id == 0){
+        variance[filter] = 0;
+        for(i = 0; i < threads; ++i){
+            variance[filter] += local[i];
+        }
+        variance[filter] /= spatial * batch;
+    }
+}
+
+extern "C" void fast_mean_gpu(float *x, int batch, int filters, int spatial, float *mean)
+{
+    fast_mean_kernel<<<filters, BLOCK>>>(x, batch, filters, spatial, mean);
+    check_error(cudaPeekAtLastError());
+}
+
+extern "C" void fast_variance_gpu(float *x, float *mean, int batch, int filters, int spatial, float *variance)
+{
+    fast_variance_kernel<<<filters, BLOCK>>>(x, mean, batch, filters, spatial, variance);
+    check_error(cudaPeekAtLastError());
+}
+
+
 extern "C" void mean_gpu(float *x, int batch, int filters, int spatial, float *mean)
 {
     mean_kernel<<<cuda_gridsize(filters), BLOCK>>>(x, batch, filters, spatial, mean);
     check_error(cudaPeekAtLastError());
 }
 
-extern "C" void fast_mean_gpu(float *x, int batch, int filters, int spatial, float *spatial_mean, float *mean)
-{
-    mean_kernel<<<cuda_gridsize(filters*batch), BLOCK>>>(x, 1, filters*batch, spatial, spatial_mean);
-    check_error(cudaPeekAtLastError());
-    mean_kernel<<<cuda_gridsize(filters), BLOCK>>>(spatial_mean, batch, filters, 1, mean);
-    check_error(cudaPeekAtLastError());
-}
-
-extern "C" void fast_variance_gpu(float *x, float *mean, int batch, int filters, int spatial, float *spatial_variance, float *variance)
-{
-    spatial_variance_kernel<<<cuda_gridsize(batch*filters), BLOCK>>>(x, mean, batch, filters, spatial, spatial_variance);
-    check_error(cudaPeekAtLastError());
-    accumulate_kernel<<<cuda_gridsize(filters), BLOCK>>>(spatial_variance, batch, filters, variance);
-    check_error(cudaPeekAtLastError());
-}
-
 extern "C" void variance_gpu(float *x, float *mean, int batch, int filters, int spatial, float *variance)
 {
     variance_kernel<<<cuda_gridsize(filters), BLOCK>>>(x, mean, batch, filters, spatial, variance);

src/classifier.c

@@ -98,6 +98,11 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile)
             sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
             save_weights(net, buff);
         }
+        if(*net.seen%1000 == 0){
+            char buff[256];
+            sprintf(buff, "%s/%s.backup",backup_directory,base);
+            save_weights(net, buff);
+        }
     }
     char buff[256];
     sprintf(buff, "%s/%s.weights", backup_directory, base);

src/coco.c

@@ -20,7 +20,8 @@ image coco_labels[80];
 void train_coco(char *cfgfile, char *weightfile)
 {
     //char *train_images = "/home/pjreddie/data/voc/test/train.txt";
-    char *train_images = "/home/pjreddie/data/coco/train.txt";
+    //char *train_images = "/home/pjreddie/data/coco/train.txt";
+    char *train_images = "data/coco.trainval.txt";
     char *backup_directory = "/home/pjreddie/backup/";
     srand(time(0));
     data_seed = time(0);

src/connected_layer.c

@@ -148,6 +148,12 @@ void forward_connected_layer_gpu(connected_layer l, network_state state)
     float * c = l.output_gpu;
     gemm_ongpu(0,0,m,n,k,1,a,k,b,n,1,c,n);
     activate_array_ongpu(l.output_gpu, l.outputs*l.batch, l.activation);
+
+/*
+    cuda_pull_array(l.output_gpu, l.output, l.outputs*l.batch);
+    float avg = mean_array(l.output, l.outputs*l.batch);
+    printf("%f\n", avg);
+    */
 }
 
 void backward_connected_layer_gpu(connected_layer l, network_state state)

src/convolutional_kernels.cu

@@ -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);
     }
 

src/convolutional_layer.c

@@ -86,9 +86,8 @@ convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int
         }
 
         l.mean = calloc(n, sizeof(float));
-        l.spatial_mean = calloc(n*l.batch, sizeof(float));
-
         l.variance = calloc(n, sizeof(float));
+
         l.rolling_mean = calloc(n, sizeof(float));
         l.rolling_variance = calloc(n, sizeof(float));
     }
@@ -114,12 +113,6 @@ convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int
         l.rolling_mean_gpu = cuda_make_array(l.mean, n);
         l.rolling_variance_gpu = cuda_make_array(l.variance, n);
 
-        l.spatial_mean_gpu = cuda_make_array(l.spatial_mean, n*l.batch);
-        l.spatial_variance_gpu = cuda_make_array(l.spatial_mean, n*l.batch);
-
-        l.spatial_mean_delta_gpu = cuda_make_array(l.spatial_mean, n*l.batch);
-        l.spatial_variance_delta_gpu = cuda_make_array(l.spatial_mean, n*l.batch);
-
         l.mean_delta_gpu = cuda_make_array(l.mean, n);
         l.variance_delta_gpu = cuda_make_array(l.variance, n);