医院做网站怎么做百度推广怎么做的

当前位置： 首页 > news >正文

医院做网站怎么做,百度推广怎么做的,做网站的底图尺寸多大,广丰网站seo算法简介 CodeFormer是一种基于AI技术深度学习的人脸复原模型#xff0c;由南洋理工大学和商汤科技联合研究中心联合开发#xff0c;它能够接收模糊或马赛克图像作为输入#xff0c;并生成更清晰的原始图像。算法源码地址#xff1a;https://github.com/sczhou/CodeFormer…算法简介 CodeFormer是一种基于AI技术深度学习的人脸复原模型由南洋理工大学和商汤科技联合研究中心联合开发它能够接收模糊或马赛克图像作为输入并生成更清晰的原始图像。算法源码地址https://github.com/sczhou/CodeFormer Face Restoration Face Color Enhancement and Restoration Face Inpainting 模型部署 如果想用C进行模型推理部署首先要把模型转换成onnx转成onnx就可以使用onnxruntime c库进行部署或者使用OpenCV的DNN也可以。 1、可在以下地址下载模型https://github.com/sczhou/CodeFormer/releases/tag/v0.1.0 2、下载CodeFormer源码在工程目录下添加onnx转换python代码 import torch from basicsr.utils.registry import ARCH_REGISTRYif name main:device torch.device(cuda:0 if torch.cuda.is_available() else cpu)net ARCH_REGISTRY.get(CodeFormer)(dim_embd512, codebook_size1024, n_head8, n_layers9, connect_list[32, 64, 128, 256]).to(device)# ckpt_path weights/CodeFormer/codeformer.pthckpt_path ./codeformer.pthcheckpoint torch.load(ckpt_path)[params_ema]net.load_state_dict(checkpoint)net.eval()input_tensor torch.zeros((1, 3, 512, 512)).to(device)torch.onnx.export(net, # 模型实例input_tensor, # 输入张量./codeformer.onnx, # 输出的ONNX模型路径export_paramsTrue, # 是否包含模型参数opset_version11, # ONNX操作集版本do_constant_foldingTrue, # 是否进行常量折叠优化input_names[input], # 输入名称output_names[output], # 输出名称dynamic_axes{input: {0: batch_size}, output: {0: batch_size}} # 声明动态轴)3、采用onnxruntime加载模型示例代码如下 #define _CRT_SECURE_NO_WARNINGS #include iostream #include fstream #include numeric #include opencv2/imgproc.hpp #include opencv2/highgui.hpp //#include cuda_provider_factory.h ///nvidia-cuda加速 #include onnxruntime_cxx_api.husing namespace cv; using namespace std; using namespace Ort;class CodeFormer { public:CodeFormer(string modelpath);Mat detect(Mat cv_image); private:void preprocess(Mat srcimg);vectorfloat inputimage;vectordouble input2_tensor;int inpWidth;int inpHeight;int outWidth;int outHeight;float min_max[2] { -1,1 };//存储初始化获得的可执行网络Env env Env(ORT_LOGGING_LEVEL_ERROR, CodeFormer);Ort::Session ort_session nullptr;SessionOptions sessionOptions SessionOptions();vectorchar input_names;vectorchar* output_names;vectorvectorint64_t input_node_dims; // 1 outputsvectorvectorint64_t output_node_dims; // 1 outputs };CodeFormer::CodeFormer(string model_path) {//OrtStatus* status OrtSessionOptionsAppendExecutionProvider_CUDA(sessionOptions, 0); ///nvidia-cuda加速sessionOptions.SetGraphOptimizationLevel(ORT_ENABLE_BASIC);std::wstring widestr std::wstring(model_path.begin(), model_path.end()); ///如果在windows系统就这么写ort_session new Session(env, widestr.c_str(), sessionOptions); ///如果在windows系统就这么写///ort_session new Session(env, model_path.c_str(), sessionOptions); ///如果在linux系统就这么写size_t numInputNodes ort_session-GetInputCount();size_t numOutputNodes ort_session-GetOutputCount();AllocatorWithDefaultOptions allocator;for (int i 0; i numInputNodes; i){input_names.push_back(ort_session-GetInputName(i, allocator));Ort::TypeInfo input_type_info ort_session-GetInputTypeInfo(i);auto input_tensor_info input_type_info.GetTensorTypeAndShapeInfo();auto input_dims input_tensor_info.GetShape();input_node_dims.push_back(input_dims);}for (int i 0; i numOutputNodes; i){output_names.push_back(ort_session-GetOutputName(i, allocator));Ort::TypeInfo output_type_info ort_session-GetOutputTypeInfo(i);auto output_tensor_info output_type_info.GetTensorTypeAndShapeInfo();auto output_dims output_tensor_info.GetShape();output_node_dims.push_back(output_dims);}this-inpHeight input_node_dims[0][2];this-inpWidth input_node_dims[0][3];this-outHeight output_node_dims[0][2];this-outWidth output_node_dims[0][3];input2_tensor.push_back(0.5); }void CodeFormer::preprocess(Mat srcimg) {Mat dstimg;cvtColor(srcimg, dstimg, COLOR_BGR2RGB);resize(dstimg, dstimg, Size(this-inpWidth, this-inpHeight), INTER_LINEAR);this-inputimage.resize(this-inpWidth * this-inpHeight * dstimg.channels());int k 0;for (int c 0; c 3; c){for (int i 0; i this-inpHeight; i){for (int j 0; j this-inpWidth; j){float pix dstimg.ptruchar(i)[j * 3 c];this-inputimagek / 0.5;k;}}} }Mat CodeFormer::detect(Mat srcimg) {int im_h srcimg.rows;int im_w srcimg.cols;this-preprocess(srcimg);arrayint64_t, 4 inputshape{ 1, 3, this-inpHeight, this-inpWidth };vectorint64_t input2shape { 1 };auto allocator_info MemoryInfo::CreateCpu(OrtDeviceAllocator, OrtMemTypeCPU);vectorValue ort_inputs;ort_inputs.push_back(Value::CreateTensorfloat(allocator_info, inputimage.data(), inputimage.size(), inputshape.data(), inputshape.size()));ort_inputs.push_back(Value::CreateTensordouble(allocator_info, input2_tensor.data(), input2_tensor.size(), input2shape.data(), input2shape.size()));vectorValue ort_outputs ort_session-Run(RunOptions{ nullptr }, input_names.data(), ort_inputs.data(), ort_inputs.size(), output_names.data(), output_names.size());post_processfloat* pred ort_outputs[0].GetTensorMutableDatafloat();//Mat mask(outHeight, outWidth, CV_32FC3, pred); /经过试验,直接这样赋值,是不行的const unsigned int channel_step outHeight * outWidth;vectorMat channel_mats;Mat rmat(outHeight, outWidth, CV_32FC1, pred); // RMat gmat(outHeight, outWidth, CV_32FC1, pred channel_step); // GMat bmat(outHeight, outWidth, CV_32FC1, pred 2 * channel_step); // Bchannel_mats.push_back(rmat);channel_mats.push_back(gmat);channel_mats.push_back(bmat);Mat mask;merge(channel_mats, mask); // CV_32FC3 allocated///不用for循环遍历Mat里的每个像素值,实现numpy.clip函数mask.setTo(this-min_max[0], mask this-min_max[0]);mask.setTo(this-min_max[1], mask this-min_max[1]); 也可以用threshold函数,阈值类型THRESH_TOZERO_INVmask (mask - this-min_max[0]) / (this-min_max[1] - this-min_max[0]);mask * 255.0;mask.convertTo(mask, CV_8UC3);cvtColor(mask, mask, COLOR_BGR2RGB);return mask; }int main() {CodeFormer mynet(codeformer.onnx);string imgpath input.png;Mat srcimg imread(imgpath);Mat dstimg mynet.detect(srcimg);resize(dstimg, dstimg, Size(srcimg.cols, srcimg.rows), INTER_LINEAR);//imwrite(result.jpg, dstimg)namedWindow(srcimg, WINDOW_NORMAL);imshow(srcimg, srcimg);namedWindow(dstimg, WINDOW_NORMAL);imshow(dstimg, dstimg);waitKey(0);destroyAllWindows(); } 效果展示 面部恢复 面部色彩增强与恢复 面部修复 破旧照片修复效果