文章详细描述了在编译和运行darknet框架时遇到的几个典型问题及解决方案，包括找不到cudnn.h头文件、CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT宏定义缺失、nvcc未找到以及不支持的GPU架构问题。解决方案包括正确配置CUDA和cudnn的路径、更新Makefile、安装libopencv-dev以及检查和设置CUDA环境变量。
问题1
设置完darknet中 Makefile中的
GPU = 1
CUDNN = 1
后执行make
编译报错：include/darknet.h:16:23: 致命错误：cudnn.h：没有那个文件或目录
解决方法：
1、找cudnn.h文件
find / -name cudnn.h
这个命令会列出所有的cudnn.h文件
找到自己安装cuda时的虚拟环境中的cudnn.h，我的是：
/root/anaconda3/envs/han/include/cudnn.h
2、把找到的cudnn.h文件复制到/usr/include
执行：cp /root/anaconda3/envs/han/include/cudnn.h /usr/include
再执行make
编译一会后仍报错报错：/usr/bin/ld: 找不到 -lcudnncollect2: 错误：ld 返回 1make: *** [libdarknet.so] 错误 1
解决方法：
1、找文件
find / -name libcudnn.so
2、复制
执行cp /root/anaconda3/envs/han/lib/libcudnn.so /usr/local/cuda-9.2/lib64
然后make，生成了可执行文件darknet和libdarknet.so及库libdarknet.a
大功告成。
注意： 其他的缺失使用同样的解决办法即可
问题2
darknet源码中CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT问题
./src/convolutional_layer.c:153:13: error: ‘CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT’ undeclared (first use in this function)
153 | CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT,
原因
cudnn8.x里是没有CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT这个宏定义的，而CUDA11.x又不能配套使用cudnn7.x，但是RTX30序列的GPU又必须使用CUDA11.x才能正常跑，感觉进了死胡同。后来找了比较久搜到NVIDIA给出了一个针对cudnn8的解决方案代码，就是修改出错的文件src/convolutional_layer.c的代码，增加针对CUDNN_MAJOR>=8的处理。
解决办法
vi src/convolutional_layer.c
替换成如下内容：
#include "convolutional_layer.h"
#include "utils.h"
#include "batchnorm_layer.h"
#include "im2col.h"
#include "col2im.h"
#include "blas.h"
#include "gemm.h"
#include <stdio.h>
#include <time.h>
#define PRINT_CUDNN_ALGO 0
#define MEMORY_LIMIT 2000000000
#ifdef AI2
#include "xnor_layer.h"
#endif
void swap_binary(convolutional_layer *l)
{
float *swap = l->weights;
l->weights = l->binary_weights;
l->binary_weights = swap;
#ifdef GPU
swap = l->weights_gpu;
l->weights_gpu = l->binary_weights_gpu;
l->binary_weights_gpu = swap;
#endif
}
void binarize_weights(float *weights, int n, int size, float *binary)
{
int i, f;
for(f = 0; f < n; ++f){
float mean = 0;
for(i = 0; i < size; ++i){
mean += fabs(weights[f*size + i]);
}
mean = mean / size;
for(i = 0; i < size; ++i){
binary[f*size + i] = (weights[f*size + i] > 0) ? mean : -mean;
}
}
}
void binarize_cpu(float *input, int n, float *binary)
{
int i;
for(i = 0; i < n; ++i){
binary[i] = (input[i] > 0) ? 1 : -1;
}
}
void binarize_input(float *input, int n, int size, float *binary)
{
int i, s;
for(s = 0; s < size; ++s){
float mean = 0;
for(i = 0; i < n; ++i){
mean += fabs(input[i*size + s]);
}
mean = mean / n;
for(i = 0; i < n; ++i){
binary[i*size + s] = (input[i*size + s] > 0) ? mean : -mean;
}
}
}
int convolutional_out_height(convolutional_layer l)
{
return (l.h + 2*l.pad - l.size) / l.stride + 1;
}
int convolutional_out_width(convolutional_layer l)
{
return (l.w + 2*l.pad - l.size) / l.stride + 1;
}
image get_convolutional_image(convolutional_layer l)
{
return float_to_image(l.out_w,l.out_h,l.out_c,l.output);
}
image get_convolutional_delta(convolutional_layer l)
{
return float_to_image(l.out_w,l.out_h,l.out_c,l.delta);
}
static size_t get_workspace_size(layer l){
#ifdef CUDNN
if(gpu_index >= 0){
size_t most = 0;
size_t s = 0;
cudnnGetConvolutionForwardWorkspaceSize(cudnn_handle(),
l.srcTensorDesc,
l.weightDesc,
l.convDesc,
l.dstTensorDesc,
l.fw_algo,
&s);
if (s > most) most = s;
cudnnGetConvolutionBackwardFilterWorkspaceSize(cudnn_handle(),
l.srcTensorDesc,
l.ddstTensorDesc,
l.convDesc,
l.dweightDesc,
l.bf_algo,
&s);
if (s > most) most = s;
cudnnGetConvolutionBackwardDataWorkspaceSize(cudnn_handle(),
l.weightDesc,
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/l.groups*sizeof(float);
}
#ifdef GPU
#ifdef CUDNN
void cudnn_convolutional_setup(layer *l)
{
cudnnSetTensor4dDescriptor(l->dsrcTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->c, l->h, l->w);
cudnnSetTensor4dDescriptor(l->ddstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->out_c, l->out_h, l->out_w);
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);
cudnnSetTensor4dDescriptor(l->normTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, 1, l->out_c, 1, 1);
cudnnSetFilter4dDescriptor(l->dweightDesc, CUDNN_DATA_FLOAT, CUDNN_TENSOR_NCHW, l->n, l->c/l->groups, l->size, l->size);
cudnnSetFilter4dDescriptor(l->weightDesc, CUDNN_DATA_FLOAT, CUDNN_TENSOR_NCHW, l->n, l->c/l->groups, l->size, l->size);
#if CUDNN_MAJOR >= 6
cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION, CUDNN_DATA_FLOAT);
#else
cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION);
#endif
#if CUDNN_MAJOR >= 7
cudnnSetConvolutionGroupCount(l->convDesc, l->groups);
#else
if(l->groups > 1){
error("CUDNN < 7 doesn't support groups, please upgrade!");
}
#endif
#if CUDNN_MAJOR >= 8
int returnedAlgoCount;
cudnnConvolutionFwdAlgoPerf_t fw_results[2 * CUDNN_CONVOLUTION_FWD_ALGO_COUNT];
cudnnConvolutionBwdDataAlgoPerf_t bd_results[2 * CUDNN_CONVOLUTION_BWD_DATA_ALGO_COUNT];
cudnnConvolutionBwdFilterAlgoPerf_t bf_results[2 * CUDNN_CONVOLUTION_BWD_FILTER_ALGO_COUNT];
cudnnFindConvolutionForwardAlgorithm(cudnn_handle(),
l->srcTensorDesc,
l->weightDesc,
l->convDesc,
l->dstTensorDesc,
CUDNN_CONVOLUTION_FWD_ALGO_COUNT,
&returnedAlgoCount,
fw_results);
for(int algoIndex = 0; algoIndex < returnedAlgoCount; ++algoIndex){
#if PRINT_CUDNN_ALGO > 0
printf("^^^^ %s for Algo %d: %f time requiring %llu memory\n",
cudnnGetErrorString(fw_results[algoIndex].status),
fw_results[algoIndex].algo, fw_results[algoIndex].time,
(unsigned long long)fw_results[algoIndex].memory);
#endif
if( fw_results[algoIndex].memory < MEMORY_LIMIT ){
l->fw_algo = fw_results[algoIndex].algo;
break;
}
}
cudnnFindConvolutionBackwardDataAlgorithm(cudnn_handle(),
l->weightDesc,
l->ddstTensorDesc,
l->convDesc,
l->dsrcTensorDesc,
CUDNN_CONVOLUTION_BWD_DATA_ALGO_COUNT,
&returnedAlgoCount,
bd_results);
for(int algoIndex = 0; algoIndex < returnedAlgoCount; ++algoIndex){
#if PRINT_CUDNN_ALGO > 0
printf("^^^^ %s for Algo %d: %f time requiring %llu memory\n",
cudnnGetErrorString(bd_results[algoIndex].status),
bd_results[algoIndex].algo, bd_results[algoIndex].time,
(unsigned long long)bd_results[algoIndex].memory);
#endif
if( bd_results[algoIndex].memory < MEMORY_LIMIT ){
l->bd_algo = bd_results[algoIndex].algo;
break;
}
}
cudnnFindConvolutionBackwardFilterAlgorithm(cudnn_handle(),
l->srcTensorDesc,
l->ddstTensorDesc,
l->convDesc,
l->dweightDesc,
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_COUNT,
&returnedAlgoCount,
bf_results);
for(int algoIndex = 0; algoIndex < returnedAlgoCount; ++algoIndex){
#if PRINT_CUDNN_ALGO > 0
printf("^^^^ %s for Algo %d: %f time requiring %llu memory\n",
cudnnGetErrorString(bf_results[algoIndex].status),
bf_results[algoIndex].algo, bf_results[algoIndex].time,
(unsigned long long)bf_results[algoIndex].memory);
#endif
if( bf_results[algoIndex].memory < MEMORY_LIMIT ){
l->bf_algo = bf_results[algoIndex].algo;
break;
}
}
#else
cudnnGetConvolutionForwardAlgorithm(cudnn_handle(),
l->srcTensorDesc,
l->weightDesc,
l->convDesc,
l->dstTensorDesc,
CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT,
2000000000,
&l->fw_algo);
cudnnGetConvolutionBackwardDataAlgorithm(cudnn_handle(),
l->weightDesc,
l->ddstTensorDesc,
l->convDesc,
l->dsrcTensorDesc,
CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT,
2000000000,
&l->bd_algo);
cudnnGetConvolutionBackwardFilterAlgorithm(cudnn_handle(),
l->srcTensorDesc,
l->ddstTensorDesc,
l->convDesc,
l->dweightDesc,
CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT,
2000000000,
&l->bf_algo);
#endif
}
#endif
#endif
convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int n, int groups, int size, int stride, int padding, ACTIVATION activation, int batch_normalize, int binary, int xnor, int adam)
{
int i;
convolutional_layer l = {0};
l.type = CONVOLUTIONAL;
l.groups = groups;
l.h = h;
l.w = w;
l.c = c;
l.n = n;
l.binary = binary;
l.xnor = xnor;
l.batch = batch;
l.stride = stride;
l.size = size;
l.pad = padding;
l.batch_normalize = batch_normalize;
l.weights = calloc(c/groups*n*size*size, sizeof(float));
l.weight_updates = calloc(c/groups*n*size*size, sizeof(float));
l.biases = calloc(n, sizeof(float));
l.bias_updates = calloc(n, sizeof(float));
l.nweights = c/groups*n*size*size;
l.nbiases = n;
// float scale = 1./sqrt(size*size*c);
float scale = sqrt(2./(size*size*c/l.groups));
//printf("convscale %f\n", scale);
//scale = .02;
//for(i = 0; i < c*n*size*size; ++i) l.weights[i] = scale*rand_uniform(-1, 1);
for(i = 0; i < l.nweights; ++i) l.weights[i] = scale*rand_normal();
int out_w = convolutional_out_width(l);
int out_h = convolutional_out_height(l);
l.out_h = out_h;
l.out_w = out_w;
l.out_c = n;
l.outputs = l.out_h * l.out_w * l.out_c;
l.inputs = l.w * l.h * l.c;
l.output = calloc(l.batch*l.outputs, sizeof(float));
l.delta = calloc(l.batch*l.outputs, sizeof(float));
l.forward = forward_convolutional_layer;
l.backward = backward_convolutional_layer;
l.update = update_convolutional_layer;
if(binary){
l.binary_weights = calloc(l.nweights, sizeof(float));
l.cweights = calloc(l.nweights, sizeof(char));
l.scales = calloc(n, sizeof(float));
}
if(xnor){
l.binary_weights = calloc(l.nweights, sizeof(float));
l.binary_input = calloc(l.inputs*l.batch, sizeof(float));
}
if(batch_normalize){
l.scales = calloc(n, sizeof(float));
l.scale_updates = calloc(n, sizeof(float));
for(i = 0; i < n; ++i){
l.scales[i] = 1;
}
l.mean = calloc(n, sizeof(float));
l.variance = calloc(n, sizeof(float));
l.mean_delta = calloc(n, sizeof(float));
l.variance_delta = calloc(n, sizeof(float));
l.rolling_mean = calloc(n, sizeof(float));
l.rolling_variance = calloc(n, sizeof(float));
l.x = calloc(l.batch*l.outputs, sizeof(float));
l.x_norm = calloc(l.batch*l.outputs, sizeof(float));
}
if(adam){
l.m = calloc(l.nweights, sizeof(float));
l.v = calloc(l.nweights, sizeof(float));
l.bias_m = calloc(n, sizeof(float));
l.scale_m = calloc(n, sizeof(float));
l.bias_v = calloc(n, sizeof(float));
l.scale_v = calloc(n, sizeof(float));
}
#ifdef GPU
l.forward_gpu = forward_convolutional_layer_gpu;
l.backward_gpu = backward_convolutional_layer_gpu;
l.update_gpu = update_convolutional_layer_gpu;
if(gpu_index >= 0){
if (adam) {
l.m_gpu = cuda_make_array(l.m, l.nweights);
l.v_gpu = cuda_make_array(l.v, l.nweights);
l.bias_m_gpu = cuda_make_array(l.bias_m, n);
l.bias_v_gpu = cuda_make_array(l.bias_v, n);
l.scale_m_gpu = cuda_make_array(l.scale_m, n);
l.scale_v_gpu = cuda_make_array(l.scale_v, n);
}
l.weights_gpu = cuda_make_array(l.weights, l.nweights);
l.weight_updates_gpu = cuda_make_array(l.weight_updates, l.nweights);
l.biases_gpu = cuda_make_array(l.biases, n);
l.bias_updates_gpu = cuda_make_array(l.bias_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);
if(binary){
l.binary_weights_gpu = cuda_make_array(l.weights, l.nweights);
}
if(xnor){
l.binary_weights_gpu = cuda_make_array(l.weights, l.nweights);
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);
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.scales_gpu = cuda_make_array(l.scales, n);
l.scale_updates_gpu = cuda_make_array(l.scale_updates, 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.normTensorDesc);
cudnnCreateTensorDescriptor(&l.srcTensorDesc);
cudnnCreateTensorDescriptor(&l.dstTensorDesc);
cudnnCreateFilterDescriptor(&l.weightDesc);
cudnnCreateTensorDescriptor(&l.dsrcTensorDesc);
cudnnCreateTensorDescriptor(&l.ddstTensorDesc);
cudnnCreateFilterDescriptor(&l.dweightDesc);
cudnnCreateConvolutionDescriptor(&l.convDesc);
cudnn_convolutional_setup(&l);
#endif
}
#endif
l.workspace_size = get_workspace_size(l);
l.activation = activation;
fprintf(stderr, "conv %5d %2d x%2d /%2d %4d x%4d x%4d -> %4d x%4d x%4d %5.3f BFLOPs\n", n, size, size, stride, w, h, c, l.out_w, l.out_h, l.out_c, (2.0 * l.n * l.size*l.size*l.c/l.groups * l.out_h*l.out_w)/1000000000.);
return l;
}
void denormalize_convolutional_layer(convolutional_layer l)
{
int i, j;
for(i = 0; i < l.n; ++i){
float scale = l.scales[i]/sqrt(l.rolling_variance[i] + .00001);
for(j = 0; j < l.c/l.groups*l.size*l.size; ++j){
l.weights[i*l.c/l.groups*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;
}
}
/*
void test_convolutional_layer()
{
convolutional_layer l = make_convolutional_layer(1, 5, 5, 3, 2, 5, 2, 1, LEAKY, 1, 0, 0, 0);
l.batch_normalize = 1;
float data[] = {1,1,1,1,1,
1,1,1,1,1,
1,1,1,1,1,
1,1,1,1,1,
1,1,1,1,1,
2,2,2,2,2,
2,2,2,2,2,
2,2,2,2,2,
2,2,2,2,2,
2,2,2,2,2,
3,3,3,3,3,
3,3,3,3,3,
3,3,3,3,3,
3,3,3,3,3,
3,3,3,3,3};
//net.input = data;
//forward_convolutional_layer(l);
}
*/
void resize_convolutional_layer(convolutional_layer *l, int w, int h)
{
l->w = w;
l->h = h;
int out_w = convolutional_out_width(*l);
int out_h = convolutional_out_height(*l);
l->out_w = out_w;
l->out_h = out_h;
l->outputs = l->out_h * l->out_w * l->out_c;
l->inputs = l->w * l->h * l->c;
l->output = realloc(l->output, l->batch*l->outputs*sizeof(float));
l->delta = realloc(l->delta, l->batch*l->outputs*sizeof(float));
if(l->batch_normalize){
l->x = realloc(l->x, l->batch*l->outputs*sizeof(float));
l->x_norm = realloc(l->x_norm, l->batch*l->outputs*sizeof(float));
}
#ifdef GPU
cuda_free(l->delta_gpu);
cuda_free(l->output_gpu);
l->delta_gpu = cuda_make_array(l->delta, l->batch*l->outputs);
l->output_gpu = cuda_make_array(l->output, l->batch*l->outputs);
if(l->batch_normalize){
cuda_free(l->x_gpu);
cuda_free(l->x_norm_gpu);
l->x_gpu = cuda_make_array(l->output, l->batch*l->outputs);
l->x_norm_gpu = cuda_make_array(l->output, l->batch*l->outputs);
}
#ifdef CUDNN
cudnn_convolutional_setup(l);
#endif
#endif
l->workspace_size = get_workspace_size(*l);
}
void add_bias(float *output, float *biases, int batch, int n, int size)
{
int i,j,b;
for(b = 0; b < batch; ++b){
for(i = 0; i < n; ++i){
for(j = 0; j < size; ++j){
output[(b*n + i)*size + j] += biases[i];
}
}
}
}
void scale_bias(float *output, float *scales, int batch, int n, int size)
{
int i,j,b;
for(b = 0; b < batch; ++b){
for(i = 0; i < n; ++i){
for(j = 0; j < size; ++j){
output[(b*n + i)*size + j] *= scales[i];
}
}
}
}
void backward_bias(float *bias_updates, float *delta, int batch, int n, int size)
{
int i,b;
for(b = 0; b < batch; ++b){
for(i = 0; i < n; ++i){
bias_updates[i] += sum_array(delta+size*(i+b*n), size);
}
}
}
void forward_convolutional_layer(convolutional_layer l, network net)
{
int i, j;
fill_cpu(l.outputs*l.batch, 0, l.output, 1);
if(l.xnor){
binarize_weights(l.weights, l.n, l.c/l.groups*l.size*l.size, l.binary_weights);
swap_binary(&l);
binarize_cpu(net.input, l.c*l.h*l.w*l.batch, l.binary_input);
net.input = l.binary_input;
}
int m = l.n/l.groups;
int k = l.size*l.size*l.c/l.groups;
int n = l.out_w*l.out_h;
for(i = 0; i < l.batch; ++i){
for(j = 0; j < l.groups; ++j){
float *a = l.weights + j*l.nweights/l.groups;
float *b = net.workspace;
float *c = l.output + (i*l.groups + j)*n*m;
float *im = net.input + (i*l.groups + j)*l.c/l.groups*l.h*l.w;
if (l.size == 1) {
b = im;
} else {
im2col_cpu(im, l.c/l.groups, l.h, l.w, l.size, l.stride, l.pad, b);
}
gemm(0,0,m,n,k,1,a,k,b,n,1,c,n);
}
}
if(l.batch_normalize){
forward_batchnorm_layer(l, net);
} else {
add_bias(l.output, l.biases, l.batch, l.n, l.out_h*l.out_w);
}
activate_array(l.output, l.outputs*l.batch, l.activation);
if(l.binary || l.xnor) swap_binary(&l);
}
void backward_convolutional_layer(convolutional_layer l, network net)
{
int i, j;
int m = l.n/l.groups;
int n = l.size*l.size*l.c/l.groups;
int k = l.out_w*l.out_h;
gradient_array(l.output, l.outputs*l.batch, l.activation, l.delta);
if(l.batch_normalize){
backward_batchnorm_layer(l, net);
} else {
backward_bias(l.bias_updates, l.delta, l.batch, l.n, k);
}
for(i = 0; i < l.batch; ++i){
for(j = 0; j < l.groups; ++j){
float *a = l.delta + (i*l.groups + j)*m*k;
float *b = net.workspace;
float *c = l.weight_updates + j*l.nweights/l.groups;
float *im = net.input + (i*l.groups + j)*l.c/l.groups*l.h*l.w;
float *imd = net.delta + (i*l.groups + j)*l.c/l.groups*l.h*l.w;
if(l.size == 1){
b = im;
} else {
im2col_cpu(im, l.c/l.groups, l.h, l.w,
l.size, l.stride, l.pad, b);
}
gemm(0,1,m,n,k,1,a,k,b,k,1,c,n);
if (net.delta) {
a = l.weights + j*l.nweights/l.groups;
b = l.delta + (i*l.groups + j)*m*k;
c = net.workspace;
if (l.size == 1) {
c = imd;
}
gemm(1,0,n,k,m,1,a,n,b,k,0,c,k);
if (l.size != 1) {
col2im_cpu(net.workspace, l.c/l.groups, l.h, l.w, l.size, l.stride, l.pad, imd);
}
}
}
}
}
void update_convolutional_layer(convolutional_layer l, update_args a)
{
float learning_rate = a.learning_rate*l.learning_rate_scale;
float momentum = a.momentum;
float decay = a.decay;
int batch = a.batch;
axpy_cpu(l.n, learning_rate/batch, l.bias_updates, 1, l.biases, 1);
scal_cpu(l.n, momentum, l.bias_updates, 1);
if(l.scales){
axpy_cpu(l.n, learning_rate/batch, l.scale_updates, 1, l.scales, 1);
scal_cpu(l.n, momentum, l.scale_updates, 1);
}
axpy_cpu(l.nweights, -decay*batch, l.weights, 1, l.weight_updates, 1);
axpy_cpu(l.nweights, learning_rate/batch, l.weight_updates, 1, l.weights, 1);
scal_cpu(l.nweights, momentum, l.weight_updates, 1);
}
image get_convolutional_weight(convolutional_layer l, int i)
{
int h = l.size;
int w = l.size;
int c = l.c/l.groups;
return float_to_image(w,h,c,l.weights+i*h*w*c);
}
void rgbgr_weights(convolutional_layer l)
{
int i;
for(i = 0; i < l.n; ++i){
image im = get_convolutional_weight(l, i);
if (im.c == 3) {
rgbgr_image(im);
}
}
}
void rescale_weights(convolutional_layer l, float scale, float trans)
{
int i;
for(i = 0; i < l.n; ++i){
image im = get_convolutional_weight(l, i);
if (im.c == 3) {
scale_image(im, scale);
float sum = sum_array(im.data, im.w*im.h*im.c);
l.biases[i] += sum*trans;
}
}
}
image *get_weights(convolutional_layer l)
{
image *weights = calloc(l.n, sizeof(image));
int i;
for(i = 0; i < l.n; ++i){
weights[i] = copy_image(get_convolutional_weight(l, i));
normalize_image(weights[i]);
/*
char buff[256];
sprintf(buff, "filter%d", i);
save_image(weights[i], buff);
*/
}
//error("hey");
return weights;
}
image *visualize_convolutional_layer(convolutional_layer l, char *window, image *prev_weights)
{
image *single_weights = get_weights(l);
show_images(single_weights, l.n, window);
image delta = get_convolutional_image(l);
image dc = collapse_image_layers(delta, 1);
char buff[256];
sprintf(buff, "%s: Output", window);
//show_image(dc, buff);
//save_image(dc, buff);
free_image(dc);
return single_weights;
}
问题3
yolov3 ./src/image_opencv.cpp:5:10: fatal error: opencv2/opencv.hpp: No such file or directory
为了实现yolo检测视频，更改Makefile里的opencv=1后，执行make出现以下报错
./src/image_opencv.cpp:5:10: fatal error: opencv2/opencv.hpp: No such
file or directory
解决办法：
sudo apt install libopencv-dev
问题4
解决nvcc找不到的问题/bin/sh:1:nvcc:not found
问题描述
在执行make指令进行编译的时候，遇到问题"/bin/sh:1:nvcc:not found"，如图所示:
解决方法
查看/usr/local/cuda/bin下是否有nvcc可执行程序，如果有则说明nvcc没有被设置为系统变量，执行如下命令
$ cd /usr/local/cuda/bin && ls
发现了nvcc确实已安装，则只需执行如下命令将其加入系统变量中:
$ sudo vi ~/.bashrc
在末尾行添加环境变量export PATH=$PATH:/usr/local/cuda/bin
这时再新建终端，然后切换到目录下进行make时就不会出现错误了。
问题5
nvcc fatal : Unsupported gpu architecture ‘compute_30‘
症结在于我的显卡不支持compute_30的GPU构架，与CUDA版本不兼容。
网上大多数的解决方法是找到makefile文件中CUDA_ARCH参数配置，将其中的 -gencode arch=compute_30,code=sm_30 \注释/删除，如图所示（图片来自网络）：
扫一扫
6312
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
