From ade27d29cbb5ada48b551b76be1512200e93b094 Mon Sep 17 00:00:00 2001 From: DefTruth <31974251+DefTruth@users.noreply.github.com> Date: Mon, 17 Jul 2023 23:06:51 +0800 Subject: [PATCH] [Sync][Internal] sync some internal features of paddle3d inference (#2118) * [Sync][Internal] sync some internal codes * [Sync][Internal] sync some internal features of paddle3d inference * [Sync][Internal] sync some internal features of paddle3d inference --- benchmark/paddlex/CMakeLists.txt | 6 +- benchmark/paddlex/README.md | 3 +- benchmark/paddlex/benchmark_gpu.sh | 10 +- benchmark/paddlex/benchmark_gpu_trt.sh | 8 +- ..._pp3d_cadnn.cc => benchmark_pp3d_caddn.cc} | 29 ++-- benchmark/paddlex/benchmark_x86.sh | 8 +- benchmark/paddlex/get_models.sh | 2 +- .../vision/perception/paddle3d/petr/petr.cc | 37 +++-- .../perception/paddle3d/petr/preprocessor.cc | 132 ++++-------------- .../perception/paddle3d/petr/preprocessor.h | 29 +++- 10 files changed, 107 insertions(+), 157 deletions(-) rename benchmark/paddlex/{benchmark_pp3d_cadnn.cc => benchmark_pp3d_caddn.cc} (78%) mode change 100755 => 100644 fastdeploy/vision/perception/paddle3d/petr/petr.cc diff --git a/benchmark/paddlex/CMakeLists.txt b/benchmark/paddlex/CMakeLists.txt index 90c2f2e74..d79ac18e1 100755 --- a/benchmark/paddlex/CMakeLists.txt +++ b/benchmark/paddlex/CMakeLists.txt @@ -21,7 +21,7 @@ add_executable(benchmark_structurev2_table ${PROJECT_SOURCE_DIR}/benchmark_struc add_executable(benchmark_structurev2_layout ${PROJECT_SOURCE_DIR}/benchmark_structurev2_layout.cc) add_executable(benchmark_ppshituv2_rec ${PROJECT_SOURCE_DIR}/benchmark_ppshituv2_rec.cc) add_executable(benchmark_ppshituv2_det ${PROJECT_SOURCE_DIR}/benchmark_ppshituv2_det.cc) -add_executable(benchmark_pp3d_cadnn ${PROJECT_SOURCE_DIR}/benchmark_pp3d_cadnn.cc) +add_executable(benchmark_pp3d_caddn ${PROJECT_SOURCE_DIR}/benchmark_pp3d_caddn.cc) add_executable(benchmark_pp3d_centerpoint ${PROJECT_SOURCE_DIR}/benchmark_pp3d_centerpoint.cc) if(UNIX AND (NOT APPLE) AND (NOT ANDROID)) @@ -35,7 +35,7 @@ if(UNIX AND (NOT APPLE) AND (NOT ANDROID)) target_link_libraries(benchmark_structurev2_layout ${FASTDEPLOY_LIBS} gflags pthread) target_link_libraries(benchmark_ppshituv2_rec ${FASTDEPLOY_LIBS} gflags pthread) target_link_libraries(benchmark_ppshituv2_det ${FASTDEPLOY_LIBS} gflags pthread) - target_link_libraries(benchmark_pp3d_cadnn ${FASTDEPLOY_LIBS} gflags pthread) + target_link_libraries(benchmark_pp3d_caddn ${FASTDEPLOY_LIBS} gflags pthread) target_link_libraries(benchmark_pp3d_centerpoint ${FASTDEPLOY_LIBS} gflags pthread) else() target_link_libraries(benchmark ${FASTDEPLOY_LIBS} gflags) @@ -48,7 +48,7 @@ else() target_link_libraries(benchmark_structurev2_layout ${FASTDEPLOY_LIBS} gflags) target_link_libraries(benchmark_ppshituv2_rec ${FASTDEPLOY_LIBS} gflags) target_link_libraries(benchmark_ppshituv2_det ${FASTDEPLOY_LIBS} gflags) - target_link_libraries(benchmark_pp3d_cadnn ${FASTDEPLOY_LIBS} gflags) + target_link_libraries(benchmark_pp3d_caddn ${FASTDEPLOY_LIBS} gflags) target_link_libraries(benchmark_pp3d_centerpoint ${FASTDEPLOY_LIBS} gflags) endif() # only for Android ADB test diff --git a/benchmark/paddlex/README.md b/benchmark/paddlex/README.md index 9b865cdbc..b73495e57 100755 --- a/benchmark/paddlex/README.md +++ b/benchmark/paddlex/README.md @@ -131,10 +131,11 @@ tar -zxvf MobileNetV3_small_x1_0.tgz ## 4. 各个硬件上的一键运行脚本 在准备好相关的环境配置和SDK后，可以使用本目录提供的脚本一键运行后的benchmark数据。 -- 获取模型和资源文件 +- 获取模型和资源文件 ```bash ./get_models.sh ``` + - 运行benchmark脚本 ```bash # x86 CPU Paddle backend fp32 diff --git a/benchmark/paddlex/benchmark_gpu.sh b/benchmark/paddlex/benchmark_gpu.sh index 92b4b33c4..bf80e3944 100755 --- a/benchmark/paddlex/benchmark_gpu.sh +++ b/benchmark/paddlex/benchmark_gpu.sh @@ -29,7 +29,7 @@ fi # PaddleSeg ./benchmark_ppseg --model OCRNet_HRNetW48 --image ppseg_cityscapes_demo_512x512.png --config_path $CONFIG_PATH -./benchmark_ppseg --model PP-LiteSeg-STDC1 --image ppseg_cityscapes_demo_512x512.png --config_path $CONFIG_PATH +./benchmark_ppseg --model PP-LiteSeg-STDC1 --image ppseg_cityscapes_demo_512x512.png --config_path $CONFIG_PATH ./benchmark_ppseg --model SegFormer-B0 --image ppseg_cityscapes_demo_512x512.png --config_path $CONFIG_PATH ./benchmark_ppseg --model PP-MobileSeg-Base --image ppseg_ade_val_512x512.png --config_path $CONFIG_PATH @@ -49,9 +49,9 @@ fi ./benchmark --model PP-Structurev2-vi-layoutxlm --shapes "1,512:1,512,4:1,512:1,512" --trt_shapes "1,512:1,512:1,512:1,512,4:1,512,4:1,512,4:1,512:1,512:1,512:1,512:1,512:1,512" --names "x_0:x_1:x_2:x_3" --dtypes "INT64:INT64:INT64:INT64" --disable_mkldnn --custom_tensor_value 0.2 --config_path $CONFIG_PATH # Paddle3D -./benchmark --model PETRv1_v99 --shapes "1,6,3,320,800:1,6,4,4" --names "images:img2lidars" --dtypes "FP32:FP32" --disable_mkldnn --config_path $CONFIG_PATH -./benchmark --model PETRv2_v99 --shapes "1,12,3,320,800:1,12,4,4:1,12" --names "images:img2lidars:timestamps" --dtypes "FP32:FP32:FP32" --disable_mkldnn --config_path $CONFIG_PATH -./benchmark_pp3d_centerpoint --model CenterPoint-Pillars-02Voxel --image paddle3d_centerpoint_n008_LIDAR_TOP__1533151603547590.pcd.bin --config_path $CONFIG_PATH -./benchmark_pp3d_cadnn --model CADNN_OCRNet-HRNetW18 --image paddle3d_cadnn_kitti_000780.png --config_path $CONFIG_PATH +./benchmark --model PETRv1_v99 --shapes "1,6,3,320,800:1,6,4,4" --names "images:img2lidars" --dtypes "FP32:FP32" --disable_mkldnn --config_path $CONFIG_PATH +./benchmark --model PETRv2_v99 --shapes "1,12,3,320,800:1,12,4,4:1,12" --names "images:img2lidars:timestamps" --dtypes "FP32:FP32:FP32" --disable_mkldnn --config_path $CONFIG_PATH +./benchmark_pp3d_centerpoint --model CenterPoint-Pillars-02Voxel --image paddle3d_centerpoint_n008_LIDAR_TOP__1533151603547590.pcd.bin --config_path $CONFIG_PATH +./benchmark_pp3d_caddn --model CADDN_OCRNet-HRNetW18 --image paddle3d_caddn_kitti_000780.png --config_path $CONFIG_PATH set +x diff --git a/benchmark/paddlex/benchmark_gpu_trt.sh b/benchmark/paddlex/benchmark_gpu_trt.sh index 4d449454c..24b9fb952 100755 --- a/benchmark/paddlex/benchmark_gpu_trt.sh +++ b/benchmark/paddlex/benchmark_gpu_trt.sh @@ -51,9 +51,9 @@ fi ./benchmark --model PP-Structurev2-vi-layoutxlm --shapes "1,512:1,512,4:1,512:1,512" --trt_shapes "1,512:1,512:1,512:1,512,4:1,512,4:1,512,4:1,512:1,512:1,512:1,512:1,512:1,512" --names "x_0:x_1:x_2:x_3" --dtypes "INT64:INT64:INT64:INT64" --disable_mkldnn --custom_tensor_value 0.2 --collect_trt_shape_by_custom_tensor_value --collect_trt_shape_by_device --config_path $CONFIG_PATH # Paddle3D -./benchmark --model PETRv1_v99 --shapes "1,6,3,320,800:1,6,4,4" --trt_shapes "1,6,3,320,800:1,6,3,320,800:1,6,3,320,800:1,6,4,4:1,6,4,4:1,6,4,4" --names "images:img2lidars" --dtypes "FP32:FP32" --disable_mkldnn --config_path $CONFIG_PATH -./benchmark --model PETRv2_v99 --shapes "1,12,3,320,800:1,12,4,4:1,12" --trt_shapes "1,12,3,320,800:1,12,3,320,800:1,12,3,320,800:1,12,4,4:1,12,4,4:1,12,4,4:1,12:1,12:1,12" --names "images:img2lidars:timestamps" --dtypes "FP32:FP32:FP32" --disable_mkldnn --config_path $CONFIG_PATH -./benchmark_pp3d_centerpoint --model CenterPoint-Pillars-02Voxel --image paddle3d_centerpoint_n008_LIDAR_TOP__1533151603547590.pcd.bin --config_path $CONFIG_PATH -./benchmark_pp3d_cadnn --model CADNN_OCRNet-HRNetW18 --image paddle3d_cadnn_kitti_000780.png --config_path $CONFIG_PATH +./benchmark --model PETRv1_v99 --shapes "1,6,3,320,800:1,6,4,4" --trt_shapes "1,6,3,320,800:1,6,3,320,800:1,6,3,320,800:1,6,4,4:1,6,4,4:1,6,4,4" --names "images:img2lidars" --dtypes "FP32:FP32" --disable_mkldnn --config_path $CONFIG_PATH +./benchmark --model PETRv2_v99 --shapes "1,12,3,320,800:1,12,4,4:1,12" --trt_shapes "1,12,3,320,800:1,12,3,320,800:1,12,3,320,800:1,12,4,4:1,12,4,4:1,12,4,4:1,12:1,12:1,12" --names "images:img2lidars:timestamps" --dtypes "FP32:FP32:FP32" --disable_mkldnn --config_path $CONFIG_PATH +./benchmark_pp3d_centerpoint --model CenterPoint-Pillars-02Voxel --image paddle3d_centerpoint_n008_LIDAR_TOP__1533151603547590.pcd.bin --config_path $CONFIG_PATH +./benchmark_pp3d_caddn --model CADDN_OCRNet-HRNetW18 --image paddle3d_caddn_kitti_000780.png --config_path $CONFIG_PATH set +x diff --git a/benchmark/paddlex/benchmark_pp3d_cadnn.cc b/benchmark/paddlex/benchmark_pp3d_caddn.cc similarity index 78% rename from benchmark/paddlex/benchmark_pp3d_cadnn.cc rename to benchmark/paddlex/benchmark_pp3d_caddn.cc index d16787bd4..f014bfcce 100644 --- a/benchmark/paddlex/benchmark_pp3d_cadnn.cc +++ b/benchmark/paddlex/benchmark_pp3d_caddn.cc @@ -56,24 +56,25 @@ int main(int argc, char* argv[]) { } if (config_info["backend"] == "paddle_trt" || config_info["backend"] == "trt") { - // use custom data to perform collect shapes. - option.trt_option.SetShape("images", {1, 3, 375, 1242}, - {1, 3, 375, 1242}, {1, 3, 375, 1242}); - option.trt_option.SetShape("trans_lidar_to_cam", {1, 4, 4}, - {1, 4, 4}, {1, 4, 4}); - option.trt_option.SetShape("trans_cam_to_img", {1, 3, 4}, - {1, 3, 4}, {1, 3, 4}); + // use custom data to perform collect shapes. + option.trt_option.SetShape("images", {1, 3, 375, 1242}, {1, 3, 375, 1242}, + {1, 3, 375, 1242}); + option.trt_option.SetShape("trans_lidar_to_cam", {1, 4, 4}, {1, 4, 4}, + {1, 4, 4}); + option.trt_option.SetShape("trans_cam_to_img", {1, 3, 4}, {1, 3, 4}, + {1, 3, 4}); std::vector image_data; - image_data.assign(im.data, im.data + 1*3*375*1242); - option.trt_option.SetInputData("trans_lidar_to_cam", lidar_data); - option.trt_option.SetInputData("trans_cam_to_img", cam_data); - option.trt_option.SetInputData("images", image_data); + image_data.assign(im.data, im.data + 1 * 3 * 375 * 1242); + option.trt_option.SetInputData("trans_lidar_to_cam", lidar_data); + option.trt_option.SetInputData("trans_cam_to_img", cam_data); + option.trt_option.SetInputData("images", image_data); } - auto model_cadnn = vision::perception::Caddn( - model_file, params_file, "", option, model_format); + auto model_cadnn = vision::perception::Caddn(model_file, params_file, "", + option, model_format); vision::PerceptionResult res; // Run profiling - BENCHMARK_MODEL(model_cadnn, model_cadnn.Predict(im, cam_data, lidar_data, &res)) + BENCHMARK_MODEL(model_cadnn, + model_cadnn.Predict(im, cam_data, lidar_data, &res)) std::cout << res.Str() << std::endl; #endif diff --git a/benchmark/paddlex/benchmark_x86.sh b/benchmark/paddlex/benchmark_x86.sh index ca668bb60..e2c7eba70 100755 --- a/benchmark/paddlex/benchmark_x86.sh +++ b/benchmark/paddlex/benchmark_x86.sh @@ -29,7 +29,7 @@ fi # PaddleSeg ./benchmark_ppseg --model OCRNet_HRNetW48 --image ppseg_cityscapes_demo_512x512.png --config_path $CONFIG_PATH -./benchmark_ppseg --model PP-LiteSeg-STDC1 --image ppseg_cityscapes_demo_512x512.png --config_path $CONFIG_PATH +./benchmark_ppseg --model PP-LiteSeg-STDC1 --image ppseg_cityscapes_demo_512x512.png --config_path $CONFIG_PATH ./benchmark_ppseg --model SegFormer-B0 --image ppseg_cityscapes_demo_512x512.png --config_path $CONFIG_PATH ./benchmark_ppseg --model PP-MobileSeg-Base --image ppseg_ade_val_512x512.png --config_path $CONFIG_PATH @@ -51,7 +51,7 @@ fi # Paddle3D ./benchmark --model PETRv1_v99 --config_path $CONFIG_PATH --shapes "1,6,3,320,800:1,6,4,4" --names "images:img2lidars" --dtypes "FP32:FP32" --disable_mkldnn --warmup 5 --repeat 20 ./benchmark --model PETRv2_v99 --config_path $CONFIG_PATH --shapes "1,12,3,320,800:1,12,4,4:1,12" --names "images:img2lidars:timestamps" --dtypes "FP32:FP32:FP32" --disable_mkldnn --warmup 5 --repeat 20 -./benchmark_pp3d_centerpoint --model CenterPoint-Pillars-02Voxel --image paddle3d_centerpoint_n008_LIDAR_TOP__1533151603547590.pcd.bin --config_path $CONFIG_PATH -./benchmark_pp3d_cadnn --model CADNN_OCRNet-HRNetW18 --image paddle3d_cadnn_kitti_000780.png --config_path $CONFIG_PATH +./benchmark_pp3d_centerpoint --model CenterPoint-Pillars-02Voxel --image paddle3d_centerpoint_n008_LIDAR_TOP__1533151603547590.pcd.bin --config_path $CONFIG_PATH +./benchmark_pp3d_caddn --model CADDN_OCRNet-HRNetW18 --image paddle3d_caddn_kitti_000780.png --config_path $CONFIG_PATH -set +x \ No newline at end of file +set +x diff --git a/benchmark/paddlex/get_models.sh b/benchmark/paddlex/get_models.sh index 48f95c1b5..d2cfa4a77 100755 --- a/benchmark/paddlex/get_models.sh +++ b/benchmark/paddlex/get_models.sh @@ -68,7 +68,7 @@ download PP-Structurev2-SLANet.tgz download PP-Structurev2-vi-layoutxlm.tgz # Paddle3D -download CADNN_OCRNet-HRNetW18.tgz +download CADDN_OCRNet-HRNetW18.tgz download CenterPoint-Pillars-02Voxel.tgz download PETRv1_v99.tgz download PETRv2_v99.tgz diff --git a/fastdeploy/vision/perception/paddle3d/petr/petr.cc b/fastdeploy/vision/perception/paddle3d/petr/petr.cc old mode 100755 new mode 100644 index 0a08b6404..770fee797 --- a/fastdeploy/vision/perception/paddle3d/petr/petr.cc +++ b/fastdeploy/vision/perception/paddle3d/petr/petr.cc @@ -19,8 +19,8 @@ namespace vision { namespace perception { Petr::Petr(const std::string& model_file, const std::string& params_file, - const std::string& config_file, const RuntimeOption& custom_option, - const ModelFormat& model_format) + const std::string& config_file, const RuntimeOption& custom_option, + const ModelFormat& model_format) : preprocessor_(config_file) { valid_cpu_backends = {Backend::PDINFER}; valid_gpu_backends = {Backend::PDINFER}; @@ -41,29 +41,38 @@ bool Petr::Initialize() { return true; } -bool Petr::Predict(const cv::Mat& im, PerceptionResult* result) { - std::vector results; - if (!BatchPredict({im}, &results)) { - return false; - } - if (results.size()) { - *result = std::move(results[0]); - } - return true; +bool Petr::Predict(const cv::Mat& images, PerceptionResult* results) { + FDERROR << "Petr inference only support 6(V1) or 12(V2) images" << std::endl; + return false; } bool Petr::BatchPredict(const std::vector& images, - std::vector* results) { + std::vector* results) { + if ((images.size() != 6) && (images.size() != 12)) { + FDERROR << "Petr only support 6(V1) or 12(V2) images"; + return false; + } std::vector fd_images = WrapMat(images); if (!preprocessor_.Run(&fd_images, &reused_input_tensors_)) { FDERROR << "Failed to preprocess the input image." << std::endl; return false; } - + + // Note: un-commented the codes below to show the debug info. + // reused_input_tensors_[0].PrintInfo(); + // reused_input_tensors_[1].PrintInfo(); + // reused_input_tensors_[2].PrintInfo(); + reused_input_tensors_[0].name = InputInfoOfRuntime(0).name; reused_input_tensors_[1].name = InputInfoOfRuntime(1).name; - reused_input_tensors_[2].name = InputInfoOfRuntime(2).name; + if (images.size() == 12) { + // for Petr V2 timestamp + reused_input_tensors_[2].name = InputInfoOfRuntime(2).name; + } else { + // for Petr V1 + reused_input_tensors_.pop_back(); + } if (!Infer(reused_input_tensors_, &reused_output_tensors_)) { FDERROR << "Failed to inference by runtime." << std::endl; diff --git a/fastdeploy/vision/perception/paddle3d/petr/preprocessor.cc b/fastdeploy/vision/perception/paddle3d/petr/preprocessor.cc index 6c3b4d5ec..61129dd52 100644 --- a/fastdeploy/vision/perception/paddle3d/petr/preprocessor.cc +++ b/fastdeploy/vision/perception/paddle3d/petr/preprocessor.cc @@ -13,6 +13,7 @@ // limitations under the License. #include "fastdeploy/vision/perception/paddle3d/petr/preprocessor.h" + #include #include "fastdeploy/function/concat.h" @@ -31,78 +32,20 @@ PetrPreprocessor::PetrPreprocessor(const std::string& config_file) { bool PetrPreprocessor::BuildPreprocessPipelineFromConfig() { processors_.clear(); - YAML::Node cfg; - try { - cfg = YAML::LoadFile(config_file_); - } catch (YAML::BadFile& e) { - FDERROR << "Failed to load yaml file " << config_file_ - << ", maybe you should check this file." << std::endl; - return false; - } - // read for preprocess - bool has_permute = false; - for (const auto& op : cfg["Preprocess"]) { - std::string op_name = op["type"].as(); - if (op_name == "NormalizeImage") { - auto mean = op["mean"].as>(); - auto std = op["std"].as>(); - bool is_scale = true; - if (op["is_scale"]) { - is_scale = op["is_scale"].as(); - } - std::string norm_type = "mean_std"; - if (op["norm_type"]) { - norm_type = op["norm_type"].as(); - } - if (norm_type != "mean_std") { - std::fill(mean.begin(), mean.end(), 0.0); - std::fill(std.begin(), std.end(), 1.0); - } - mean_ = mean; - std_ = std; - } else if (op_name == "Resize") { - bool keep_ratio = op["keep_ratio"].as(); - auto target_size = op["target_size"].as>(); - int interp = op["interp"].as(); - FDASSERT(target_size.size() == 2, - "Require size of target_size be 2, but now it's %lu.", - target_size.size()); - if (!keep_ratio) { - int width = target_size[0]; - int height = target_size[1]; - processors_.push_back( - std::make_shared(width, height, -1.0, -1.0, interp, false)); - } else { - int min_target_size = std::min(target_size[0], target_size[1]); - int max_target_size = std::max(target_size[0], target_size[1]); - std::vector max_size; - if (max_target_size > 0) { - max_size.push_back(max_target_size); - max_size.push_back(max_target_size); - } - processors_.push_back(std::make_shared( - min_target_size, interp, true, max_size)); - } - } else if (op_name == "Permute") { - // Do nothing, do permute as the last operation - has_permute = true; - continue; - } else { - FDERROR << "Unexcepted preprocess operator: " << op_name << "." - << std::endl; - return false; - } - } - if (!disable_permute_) { - if (has_permute) { - // permute = cast + HWC2CHW - processors_.push_back(std::make_shared("float")); - processors_.push_back(std::make_shared()); - } - } + processors_.push_back(std::make_shared(800, 450)); + processors_.push_back(std::make_shared(0, 130, 800, 320)); + + std::vector mean{103.530, 116.280, 123.675}; + std::vector std{57.375, 57.120, 58.395}; + bool scale = false; + processors_.push_back(std::make_shared(mean, std, scale)); + processors_.push_back(std::make_shared("float")); + processors_.push_back(std::make_shared()); + + // Fusion will improve performance + FuseTransforms(&processors_); - input_k_data_ = cfg["k_data"].as>(); return true; } @@ -119,16 +62,16 @@ bool PetrPreprocessor::Apply(FDMatBatch* image_batch, } // There are 3 outputs, image, k_data, timestamp outputs->resize(3); - int batch = static_cast(image_batch->mats->size()); + int num_cams = static_cast(image_batch->mats->size()); // Allocate memory for k_data - (*outputs)[1].Resize({1, batch, 4, 4}, FDDataType::FP32); + (*outputs)[1].Resize({1, num_cams, 4, 4}, FDDataType::FP32); // Allocate memory for image_data - (*outputs)[0].Resize({1, batch, 3, 320, 800}, FDDataType::FP32); + (*outputs)[0].Resize({1, num_cams, 3, 320, 800}, FDDataType::FP32); // Allocate memory for timestamp - (*outputs)[2].Resize({1, batch}, FDDataType::FP32); + (*outputs)[2].Resize({1, num_cams}, FDDataType::FP32); auto* image_ptr = reinterpret_cast((*outputs)[0].MutableData()); @@ -144,53 +87,28 @@ bool PetrPreprocessor::Apply(FDMatBatch* image_batch, << processors_[j]->Name() << "." << std::endl; return false; } - if (processors_[j]->Name() == "Resize") { - // crop and normalize after Resize - auto img = *(mat->GetOpenCVMat()); - cv::Mat crop_img = img(cv::Range(130, 450), cv::Range(0, 800)); - Normalize(&crop_img, mean_, std_, scale_); - FDMat fd_mat = WrapMat(crop_img); - image_batch->mats->at(i) = fd_mat; - } } } - for (int i = 0; i < batch / 2 * 4 * 4; ++i) { - input_k_data_.emplace_back(input_k_data_[i]); + for (int i = 0; i < num_cams / 2 * 4 * 4; ++i) { + input_k_data_.push_back(input_k_data_[i]); } + memcpy(k_data_ptr, input_k_data_.data(), num_cams * 16 * sizeof(float)); - memcpy(k_data_ptr, input_k_data_.data(), batch * 16 * sizeof(float)); - - std::vector timestamp(batch, 0.0f); - for (int i = batch / 2; i < batch; ++i) { + std::vector timestamp(num_cams, 0.0f); + for (int i = num_cams / 2; i < num_cams; ++i) { timestamp[i] = 1.0f; } - memcpy(timestamp_ptr, timestamp.data(), batch * sizeof(float)); + memcpy(timestamp_ptr, timestamp.data(), num_cams * sizeof(float)); - FDTensor* tensor = image_batch->Tensor(); + FDTensor* tensor = image_batch->Tensor(); // [num_cams,3,320,800] + tensor->ExpandDim(0); // [num_cams,3,320,800] -> [1,num_cams,3,320,800] (*outputs)[0].SetExternalData(tensor->Shape(), tensor->Dtype(), tensor->Data(), tensor->device, tensor->device_id); return true; } -void PetrPreprocessor::Normalize(cv::Mat* im, const std::vector& mean, - const std::vector& std, float& scale) { - if (scale) { - (*im).convertTo(*im, CV_32FC3, scale); - } - for (int h = 0; h < im->rows; h++) { - for (int w = 0; w < im->cols; w++) { - im->at(h, w)[0] = - (im->at(h, w)[0] - mean[0]) / std[0]; - im->at(h, w)[1] = - (im->at(h, w)[1] - mean[1]) / std[1]; - im->at(h, w)[2] = - (im->at(h, w)[2] - mean[2]) / std[2]; - } - } -} - } // namespace perception } // namespace vision } // namespace fastdeploy diff --git a/fastdeploy/vision/perception/paddle3d/petr/preprocessor.h b/fastdeploy/vision/perception/paddle3d/petr/preprocessor.h index 381c81abe..320140026 100755 --- a/fastdeploy/vision/perception/paddle3d/petr/preprocessor.h +++ b/fastdeploy/vision/perception/paddle3d/petr/preprocessor.h @@ -41,9 +41,6 @@ class FASTDEPLOY_DECL PetrPreprocessor : public ProcessorManager { */ bool Apply(FDMatBatch* image_batch, std::vector* outputs); - void Normalize(cv::Mat *im, const std::vector &mean, - const std::vector &std, float &scale); - protected: bool BuildPreprocessPipelineFromConfig(); std::vector> processors_; @@ -58,7 +55,31 @@ class FASTDEPLOY_DECL PetrPreprocessor : public ProcessorManager { std::vector mean_; std::vector std_; - std::vector input_k_data_; + std::vector input_k_data_{ + -1.40307297e-03, 9.07780395e-06, 4.84838307e-01, -5.43047376e-02, + -1.40780103e-04, 1.25770375e-05, 1.04126692e+00, 7.67668605e-01, + -1.02884378e-05, -1.41007011e-03, 1.02823459e-01, -3.07415128e-01, + 0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 1.00000000e+00, + -9.39000631e-04, -7.65239349e-07, 1.14073277e+00, 4.46270645e-01, + 1.04998052e-03, 1.91798881e-05, 2.06218868e-01, 7.42717385e-01, + 1.48074005e-05, -1.40855671e-03, 7.45946690e-02, -3.16081315e-01, + 0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 1.00000000e+00, + -7.0699735e-04, 4.2389297e-07, -5.5183989e-01, -5.3276348e-01, + -1.2281288e-03, 2.5626015e-05, 1.0212017e+00, 6.1102939e-01, + -2.2421273e-05, -1.4170362e-03, 9.3639769e-02, -3.0863306e-01, + 0.0000000e+00, 0.0000000e+00, 0.0000000e+00, 1.0000000e+00, + 2.2227580e-03, 2.5312484e-06, -9.7261822e-01, 9.0684637e-02, + 1.9360810e-04, 2.1347081e-05, -1.0779887e+00, -7.9227984e-01, + 4.3742721e-06, -2.2310747e-03, 1.0842450e-01, -2.9406491e-01, + 0.0000000e+00, 0.0000000e+00, 0.0000000e+00, 1.0000000e+00, + 5.97175560e-04, -5.88774265e-06, -1.15893924e+00, -4.49921310e-01, + -1.28312141e-03, 3.58297058e-07, 1.48300052e-01, 1.14334166e-01, + -2.80917516e-06, -1.41527120e-03, 8.37693438e-02, -2.36765608e-01, + 0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 1.00000000e+00, + 3.6048229e-04, 3.8333174e-06, 7.9871160e-01, 4.3321830e-01, + 1.3671946e-03, 6.7484652e-06, -8.4722507e-01, 1.9411178e-01, + 7.5027779e-06, -1.4139183e-03, 8.2083985e-02, -2.4505949e-01, + 0.0000000e+00, 0.0000000e+00, 0.0000000e+00, 1.0000000e+00}; }; } // namespace perception