haotian/duan8
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

yolov8n (#1387)

Browse Source 
* 修改yolov8nchannel数量以及 coco_file引用传递

* 优化代码风格

* 优化代码风格
This commit is contained in:
Rex 2023-10-23 10:56:10 +08:00 committed by GitHub
parent 6a267e0413
commit c997e35710
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 15 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
yolov8/src/model.cpp
View File

@ -123,7 +123,7 @@ nvinfer1::IHostMemory* buildEngineYolov8Det(nvinfer1::IBuilder* builder,
********************************************* YOLOV8 OUTPUT ******************************************
*******************************************************************************************************/
int base_in_channel = (gw == 1.25) ? 80 : 64;
int base_out_channel = (gw == 0.25) ? 320 : 256;
int base_out_channel = (gw == 0.25) ? std::max(64, std::min(kNumClass, 100)) : get_width(256, gw, max_channels);
// output0
nvinfer1::IElementWiseLayer* conv22_cv2_0_0 = convBnSiLU(network, weightMap, *conv15->getOutput(0), base_in_channel, 3, 1, 1, "model.22.cv2.0.0");
@ -131,8 +131,8 @@ nvinfer1::IHostMemory* buildEngineYolov8Det(nvinfer1::IBuilder* builder,
nvinfer1::IConvolutionLayer* conv22_cv2_0_2 = network->addConvolutionNd(*conv22_cv2_0_1->getOutput(0), 64, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv2.0.2.weight"], weightMap["model.22.cv2.0.2.bias"]);
conv22_cv2_0_2->setStrideNd(nvinfer1::DimsHW{1, 1});
conv22_cv2_0_2->setPaddingNd(nvinfer1::DimsHW{0, 0});
nvinfer1::IElementWiseLayer* conv22_cv3_0_0 = convBnSiLU(network, weightMap, *conv15->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.0.0");
nvinfer1::IElementWiseLayer* conv22_cv3_0_1 = convBnSiLU(network, weightMap, *conv22_cv3_0_0->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.0.1");
nvinfer1::IElementWiseLayer* conv22_cv3_0_0 = convBnSiLU(network, weightMap, *conv15->getOutput(0),base_out_channel, 3, 1, 1, "model.22.cv3.0.0");
nvinfer1::IElementWiseLayer* conv22_cv3_0_1 = convBnSiLU(network, weightMap, *conv22_cv3_0_0->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.0.1");
nvinfer1::IConvolutionLayer* conv22_cv3_0_2 = network->addConvolutionNd(*conv22_cv3_0_1->getOutput(0), kNumClass, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv3.0.2.weight"], weightMap["model.22.cv3.0.2.bias"]);
conv22_cv3_0_2->setStride(nvinfer1::DimsHW{1, 1});
conv22_cv3_0_2->setPadding(nvinfer1::DimsHW{0, 0});
@ -145,8 +145,8 @@ nvinfer1::IHostMemory* buildEngineYolov8Det(nvinfer1::IBuilder* builder,
nvinfer1::IConvolutionLayer* conv22_cv2_1_2 = network->addConvolutionNd(*conv22_cv2_1_1->getOutput(0), 64, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv2.1.2.weight"], weightMap["model.22.cv2.1.2.bias"]);
conv22_cv2_1_2->setStrideNd(nvinfer1::DimsHW{1, 1});
conv22_cv2_1_2->setPaddingNd(nvinfer1::DimsHW{0, 0});
nvinfer1::IElementWiseLayer* conv22_cv3_1_0 = convBnSiLU(network, weightMap, *conv18->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.1.0");
nvinfer1::IElementWiseLayer* conv22_cv3_1_1 = convBnSiLU(network, weightMap, *conv22_cv3_1_0->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.1.1");
nvinfer1::IElementWiseLayer* conv22_cv3_1_0 = convBnSiLU(network, weightMap, *conv18->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.1.0");
nvinfer1::IElementWiseLayer* conv22_cv3_1_1 = convBnSiLU(network, weightMap, *conv22_cv3_1_0->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.1.1");
nvinfer1::IConvolutionLayer* conv22_cv3_1_2 = network->addConvolutionNd(*conv22_cv3_1_1->getOutput(0), kNumClass, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv3.1.2.weight"], weightMap["model.22.cv3.1.2.bias"]);
conv22_cv3_1_2->setStrideNd(nvinfer1::DimsHW{1, 1});
conv22_cv3_1_2->setPaddingNd(nvinfer1::DimsHW{0, 0});
@ -157,8 +157,8 @@ nvinfer1::IHostMemory* buildEngineYolov8Det(nvinfer1::IBuilder* builder,
nvinfer1::IElementWiseLayer* conv22_cv2_2_0 = convBnSiLU(network, weightMap, *conv21->getOutput(0), base_in_channel, 3, 1, 1, "model.22.cv2.2.0");
nvinfer1::IElementWiseLayer* conv22_cv2_2_1 = convBnSiLU(network, weightMap, *conv22_cv2_2_0->getOutput(0), base_in_channel, 3, 1, 1, "model.22.cv2.2.1");
nvinfer1::IConvolutionLayer* conv22_cv2_2_2 = network->addConvolution(*conv22_cv2_2_1->getOutput(0), 64, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv2.2.2.weight"], weightMap["model.22.cv2.2.2.bias"]);
nvinfer1::IElementWiseLayer* conv22_cv3_2_0 = convBnSiLU(network, weightMap, *conv21->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.2.0");
nvinfer1::IElementWiseLayer* conv22_cv3_2_1 = convBnSiLU(network, weightMap, *conv22_cv3_2_0->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.2.1");
nvinfer1::IElementWiseLayer* conv22_cv3_2_0 = convBnSiLU(network, weightMap, *conv21->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.2.0");
nvinfer1::IElementWiseLayer* conv22_cv3_2_1 = convBnSiLU(network, weightMap, *conv22_cv3_2_0->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.2.1");
nvinfer1::IConvolutionLayer* conv22_cv3_2_2 = network->addConvolution(*conv22_cv3_2_1->getOutput(0), kNumClass, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv3.2.2.weight"], weightMap["model.22.cv3.2.2.bias"]);
nvinfer1::ITensor* inputTensor22_2[] = {conv22_cv2_2_2->getOutput(0), conv22_cv3_2_2->getOutput(0)};
nvinfer1::IConcatenationLayer* cat22_2 = network->addConcatenation(inputTensor22_2, 2);
@ -281,7 +281,7 @@ nvinfer1::IHostMemory* buildEngineYolov8Seg(nvinfer1::IBuilder* builder,
********************************************* YOLOV8 OUTPUT ******************************************
*******************************************************************************************************/
int base_in_channel = (gw == 1.25) ? 80 : 64;
int base_out_channel = (gw == 0.25) ? 320 : 256;
int base_out_channel = (gw == 0.25) ? std::max(64, std::min(kNumClass, 100)) : get_width(256, gw, max_channels);
// output0
nvinfer1::IElementWiseLayer* conv22_cv2_0_0 = convBnSiLU(network, weightMap, *conv15->getOutput(0), base_in_channel, 3, 1, 1, "model.22.cv2.0.0");
@ -289,8 +289,8 @@ nvinfer1::IHostMemory* buildEngineYolov8Seg(nvinfer1::IBuilder* builder,
nvinfer1::IConvolutionLayer* conv22_cv2_0_2 = network->addConvolutionNd(*conv22_cv2_0_1->getOutput(0), 64, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv2.0.2.weight"], weightMap["model.22.cv2.0.2.bias"]);
conv22_cv2_0_2->setStrideNd(nvinfer1::DimsHW{1, 1});
conv22_cv2_0_2->setPaddingNd(nvinfer1::DimsHW{0, 0});
nvinfer1::IElementWiseLayer *conv22_cv3_0_0 = convBnSiLU(network, weightMap, *conv15->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.0.0");
nvinfer1::IElementWiseLayer *conv22_cv3_0_1 = convBnSiLU(network, weightMap, *conv22_cv3_0_0->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.0.1");
nvinfer1::IElementWiseLayer *conv22_cv3_0_0 = convBnSiLU(network, weightMap, *conv15->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.0.0");
nvinfer1::IElementWiseLayer *conv22_cv3_0_1 = convBnSiLU(network, weightMap, *conv22_cv3_0_0->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.0.1");
nvinfer1::IConvolutionLayer *conv22_cv3_0_2 = network->addConvolutionNd(*conv22_cv3_0_1->getOutput(0), kNumClass, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv3.0.2.weight"], weightMap["model.22.cv3.0.2.bias"]);
conv22_cv3_0_2->setStride(nvinfer1::DimsHW{1, 1});
conv22_cv3_0_2->setPadding(nvinfer1::DimsHW{0, 0});
@ -303,8 +303,8 @@ nvinfer1::IHostMemory* buildEngineYolov8Seg(nvinfer1::IBuilder* builder,
nvinfer1::IConvolutionLayer* conv22_cv2_1_2 = network->addConvolutionNd(*conv22_cv2_1_1->getOutput(0), 64, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv2.1.2.weight"], weightMap["model.22.cv2.1.2.bias"]);
conv22_cv2_1_2->setStrideNd(nvinfer1::DimsHW{1, 1});
conv22_cv2_1_2->setPaddingNd(nvinfer1::DimsHW{0, 0});
nvinfer1::IElementWiseLayer* conv22_cv3_1_0 = convBnSiLU(network, weightMap, *conv18->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.1.0");
nvinfer1::IElementWiseLayer* conv22_cv3_1_1 = convBnSiLU(network, weightMap, *conv22_cv3_1_0->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.1.1");
nvinfer1::IElementWiseLayer* conv22_cv3_1_0 = convBnSiLU(network, weightMap, *conv18->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.1.0");
nvinfer1::IElementWiseLayer* conv22_cv3_1_1 = convBnSiLU(network, weightMap, *conv22_cv3_1_0->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.1.1");
nvinfer1::IConvolutionLayer* conv22_cv3_1_2 = network->addConvolutionNd(*conv22_cv3_1_1->getOutput(0), kNumClass, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv3.1.2.weight"], weightMap["model.22.cv3.1.2.bias"]);
conv22_cv3_1_2->setStrideNd(nvinfer1::DimsHW{1, 1});
conv22_cv3_1_2->setPaddingNd(nvinfer1::DimsHW{0, 0});
@ -315,8 +315,8 @@ nvinfer1::IHostMemory* buildEngineYolov8Seg(nvinfer1::IBuilder* builder,
nvinfer1::IElementWiseLayer* conv22_cv2_2_0 = convBnSiLU(network, weightMap, *conv21->getOutput(0), base_in_channel, 3, 1, 1, "model.22.cv2.2.0");
nvinfer1::IElementWiseLayer* conv22_cv2_2_1 = convBnSiLU(network, weightMap, *conv22_cv2_2_0->getOutput(0), base_in_channel, 3, 1, 1, "model.22.cv2.2.1");
nvinfer1::IConvolutionLayer* conv22_cv2_2_2 = network->addConvolution(*conv22_cv2_2_1->getOutput(0), 64, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv2.2.2.weight"], weightMap["model.22.cv2.2.2.bias"]);
nvinfer1::IElementWiseLayer* conv22_cv3_2_0 = convBnSiLU(network, weightMap, *conv21->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.2.0");
nvinfer1::IElementWiseLayer* conv22_cv3_2_1 = convBnSiLU(network, weightMap, *conv22_cv3_2_0->getOutput(0), get_width(base_out_channel, gw, max_channels), 3, 1, 1, "model.22.cv3.2.1");
nvinfer1::IElementWiseLayer* conv22_cv3_2_0 = convBnSiLU(network, weightMap, *conv21->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.2.0");
nvinfer1::IElementWiseLayer* conv22_cv3_2_1 = convBnSiLU(network, weightMap, *conv22_cv3_2_0->getOutput(0), base_out_channel, 3, 1, 1, "model.22.cv3.2.1");
nvinfer1::IConvolutionLayer* conv22_cv3_2_2 = network->addConvolution(*conv22_cv3_2_1->getOutput(0), kNumClass, nvinfer1::DimsHW{1, 1}, weightMap["model.22.cv3.2.2.weight"], weightMap["model.22.cv3.2.2.bias"]);
nvinfer1::ITensor* inputTensor22_2[] = {conv22_cv2_2_2->getOutput(0), conv22_cv3_2_2->getOutput(0)};
nvinfer1::IConcatenationLayer* cat22_2 = network->addConcatenation(inputTensor22_2, 2);

2
yolov8/yolov8_seg.cpp
View File

@ -163,7 +163,7 @@ void infer(IExecutionContext &context, cudaStream_t &stream, void **buffers, flo
}
bool parse_args(int argc, char **argv, std::string &wts, std::string &engine, std::string &img_dir, std::string &sub_type,
std::string &cuda_post_process, std::string labels_filename, float &gd, float &gw, int &max_channels)
std::string &cuda_post_process, std::string& labels_filename, float &gd, float &gw, int &max_channels)
{
if (argc < 4)
return false;