mmdeploy/csrc/core/model.h

154 lines
4.0 KiB
C
Raw Normal View History

Merge sdk (#251) * check in cmake * move backend_ops to csrc/backend_ops * check in preprocess, model, some codebase and their c-apis * check in CMakeLists.txt * check in parts of test_csrc * commit everything else * add readme * update core's BUILD_INTERFACE directory * skip codespell on third_party * update trt_net and ort_net's CMakeLists * ignore clion's build directory * check in pybind11 * add onnx.proto. Remove MMDeploy's dependency on ncnn's source code * export MMDeployTargets only when MMDEPLOY_BUILD_SDK is ON * remove useless message * target include directory is wrong * change target name from mmdeploy_ppl_net to mmdeploy_pplnn_net * skip install directory * update project's cmake * remove useless code * set CMAKE_BUILD_TYPE to Release by force if it isn't set by user * update custom ops CMakeLists * pass object target's source lists * fix lint end-of-file * fix lint: trailing whitespace * fix codespell hook * remove bicubic_interpolate to csrc/backend_ops/ * set MMDEPLOY_BUILD_SDK OFF * change custom ops build command * add spdlog installation command * update docs on how to checkout pybind11 * move bicubic_interpolate to backend_ops/tensorrt directory * remove useless code * correct cmake * fix typo * fix typo * fix install directory * correct sdk's readme * set cub dir when cuda version < 11.0 * change directory where clang-format will apply to * fix build command * add .clang-format * change clang-format style from google to file * reformat csrc/backend_ops * format sdk's code * turn off clang-format for some files * add -Xcompiler=-fno-gnu-unique * fix trt topk initialize * check in config for sdk demo * update cmake script and csrc's readme * correct config's path * add cuda include directory, otherwise compile failed in case of tensorrt8.2 * clang-format onnx2ncnn.cpp Co-authored-by: zhangli <lzhang329@gmail.com> Co-authored-by: grimoire <yaoqian@sensetime.com>
2021-12-07 10:57:55 +08:00
// Copyright (c) OpenMMLab. All rights reserved.
#ifndef CORE_SDK_MODEL_H
#define CORE_SDK_MODEL_H
#include <functional>
#include <memory>
#include <string>
#include <vector>
#include "serialization.h"
#include "types.h"
namespace mmdeploy {
struct model_meta_info_t {
std::string name;
std::string net;
std::string weights;
std::string backend;
int batch_size;
std::string precision;
bool dynamic_shape;
MMDEPLOY_ARCHIVE_MEMBERS(name, net, weights, backend, batch_size, precision, dynamic_shape);
};
struct deploy_meta_info_t {
std::string version;
std::vector<model_meta_info_t> models;
MMDEPLOY_ARCHIVE_MEMBERS(version, models);
Merge sdk (#251) * check in cmake * move backend_ops to csrc/backend_ops * check in preprocess, model, some codebase and their c-apis * check in CMakeLists.txt * check in parts of test_csrc * commit everything else * add readme * update core's BUILD_INTERFACE directory * skip codespell on third_party * update trt_net and ort_net's CMakeLists * ignore clion's build directory * check in pybind11 * add onnx.proto. Remove MMDeploy's dependency on ncnn's source code * export MMDeployTargets only when MMDEPLOY_BUILD_SDK is ON * remove useless message * target include directory is wrong * change target name from mmdeploy_ppl_net to mmdeploy_pplnn_net * skip install directory * update project's cmake * remove useless code * set CMAKE_BUILD_TYPE to Release by force if it isn't set by user * update custom ops CMakeLists * pass object target's source lists * fix lint end-of-file * fix lint: trailing whitespace * fix codespell hook * remove bicubic_interpolate to csrc/backend_ops/ * set MMDEPLOY_BUILD_SDK OFF * change custom ops build command * add spdlog installation command * update docs on how to checkout pybind11 * move bicubic_interpolate to backend_ops/tensorrt directory * remove useless code * correct cmake * fix typo * fix typo * fix install directory * correct sdk's readme * set cub dir when cuda version < 11.0 * change directory where clang-format will apply to * fix build command * add .clang-format * change clang-format style from google to file * reformat csrc/backend_ops * format sdk's code * turn off clang-format for some files * add -Xcompiler=-fno-gnu-unique * fix trt topk initialize * check in config for sdk demo * update cmake script and csrc's readme * correct config's path * add cuda include directory, otherwise compile failed in case of tensorrt8.2 * clang-format onnx2ncnn.cpp Co-authored-by: zhangli <lzhang329@gmail.com> Co-authored-by: grimoire <yaoqian@sensetime.com>
2021-12-07 10:57:55 +08:00
};
class ModelImpl;
/**
* @class Model
* @brief Read sdk model from file.
* @note there might be more than one models in an sdk model file. For example,
* in case of faster-rcnn model, it splits into two models, one is rpn and the
* other is cnn for roi classification.
*/
Support Windows (#106) * minor changes * support windows * fix GCC build * fix lint * reformat * fix Windows build * fix GCC build * search backend ops for onnxruntime * fix lint * fix lint * code clean-up * code clean-up * fix clang build * fix trt support * fix cmake for ncnn * fix cmake for openvino * fix SDK Python API * handle ops for other backends (ncnn, trt) * handle SDK Python API library location * robustify linkage * fix cuda * minor fix for openvino & ncnn * use CMAKE_CUDA_ARCHITECTURES if set * fix cuda preprocessor * fix misc * fix pplnn & pplcv, drop support for pplcv<0.6.0 * robustify cmake * update build.md (#2) * build dynamic modules as module library & fix demo (partially) * fix candidate path for mmdeploy_python * move "enable CUDA" to cmake config for demo * refine demo cmake * add comment * fix ubuntu build * revert docs/en/build.md * fix C API * fix lint * Windows build doc (#3) * check in docs related to mmdeploy build on windows * update build guide on windows platform * update build guide on windows platform * make path of thirdparty libraries consistent * make path consistency * correct build command for custom ops * correct build command for sdk * update sdk build instructions * update doc * correct build command * fix lint * correct build command and fix lint Co-authored-by: lvhan <lvhan@pjlab.org> * trailing whitespace (#4) * minor fix * fix sr sdk model * fix type deduction * fix cudaFree after driver shutting down * update ppl.cv installation warning (#5) * fix device allocator threshold & fix lint * update doc (#6) * update ppl.cv installation warning * missing 'git clone' Co-authored-by: chenxin <chenxin2@sensetime.com> Co-authored-by: zhangli <zhangli@sensetime.com> Co-authored-by: lvhan028 <lvhan_028@163.com> Co-authored-by: lvhan <lvhan@pjlab.org>
2022-02-24 20:08:44 +08:00
class MMDEPLOY_API Model {
Merge sdk (#251) * check in cmake * move backend_ops to csrc/backend_ops * check in preprocess, model, some codebase and their c-apis * check in CMakeLists.txt * check in parts of test_csrc * commit everything else * add readme * update core's BUILD_INTERFACE directory * skip codespell on third_party * update trt_net and ort_net's CMakeLists * ignore clion's build directory * check in pybind11 * add onnx.proto. Remove MMDeploy's dependency on ncnn's source code * export MMDeployTargets only when MMDEPLOY_BUILD_SDK is ON * remove useless message * target include directory is wrong * change target name from mmdeploy_ppl_net to mmdeploy_pplnn_net * skip install directory * update project's cmake * remove useless code * set CMAKE_BUILD_TYPE to Release by force if it isn't set by user * update custom ops CMakeLists * pass object target's source lists * fix lint end-of-file * fix lint: trailing whitespace * fix codespell hook * remove bicubic_interpolate to csrc/backend_ops/ * set MMDEPLOY_BUILD_SDK OFF * change custom ops build command * add spdlog installation command * update docs on how to checkout pybind11 * move bicubic_interpolate to backend_ops/tensorrt directory * remove useless code * correct cmake * fix typo * fix typo * fix install directory * correct sdk's readme * set cub dir when cuda version < 11.0 * change directory where clang-format will apply to * fix build command * add .clang-format * change clang-format style from google to file * reformat csrc/backend_ops * format sdk's code * turn off clang-format for some files * add -Xcompiler=-fno-gnu-unique * fix trt topk initialize * check in config for sdk demo * update cmake script and csrc's readme * correct config's path * add cuda include directory, otherwise compile failed in case of tensorrt8.2 * clang-format onnx2ncnn.cpp Co-authored-by: zhangli <lzhang329@gmail.com> Co-authored-by: grimoire <yaoqian@sensetime.com>
2021-12-07 10:57:55 +08:00
public:
Model() = default;
/**
* @brief construct `Model` with an sdk model's path
* @param model_path file path of an sdk model. It can be a file or a
* directory. Refer to `Load`
* @note An exception might be threw. `Try...catch...` is strongly recommended
* when this constructor is used
*/
explicit Model(const std::string& model_path);
Model(const void* buffer, size_t size);
~Model() = default;
/**
* @brief Load an sdk model.
* @param model_path file path of an sdk model. It can be a file or a
* directory.
* @return status with an error code.
*/
Result<void> Init(const std::string& model_path);
Result<void> Init(const void* buffer, size_t size);
/**
* @brief Return a specified model's meta info
* @param name the name of a model in sdk model file
* @return
*/
Result<model_meta_info_t> GetModelConfig(const std::string& name) const;
/**
* @brief Read specified file from an sdk model
* @param file_path path relative to the root directory of an sdk model.
* @return the content of specified file if success, which can be accessed by
* `Result<T>.value()`. Otherwise, error code is returned that can be obtained
* by `Result<T>.error()`
*/
Result<std::string> ReadFile(const std::string& file_path) noexcept;
/**
* @brief get meta information of an sdk model
* @return sdk model's meta information
*/
const deploy_meta_info_t& meta() const { return meta_; }
/**
* @brief Check if an instance of `Model` is valid
* @return the status of an instance of `Model`
*/
explicit operator bool() const { return impl_ != nullptr; }
private:
std::shared_ptr<ModelImpl> impl_;
deploy_meta_info_t meta_;
};
/**
* @class ModelRegistry
* @brief SDK model implementor's factory. The following code shows how to
* register a new implementor to the factory.
* @example
* class ANewModelImpl : public ModelImpl {
* };
* class ANewModelImplRegister {
* public:
* ANewModelImplRegister() {
* ModelRegistry::Get().Register("ANewModelImpl",
* []()->unique_ptr<ModelImpl>{return make_unique<ANewModelImpl>();});
* }
* };
* ANewModelImplRegister a_new_model_impl_register;
*/
Support Windows (#106) * minor changes * support windows * fix GCC build * fix lint * reformat * fix Windows build * fix GCC build * search backend ops for onnxruntime * fix lint * fix lint * code clean-up * code clean-up * fix clang build * fix trt support * fix cmake for ncnn * fix cmake for openvino * fix SDK Python API * handle ops for other backends (ncnn, trt) * handle SDK Python API library location * robustify linkage * fix cuda * minor fix for openvino & ncnn * use CMAKE_CUDA_ARCHITECTURES if set * fix cuda preprocessor * fix misc * fix pplnn & pplcv, drop support for pplcv<0.6.0 * robustify cmake * update build.md (#2) * build dynamic modules as module library & fix demo (partially) * fix candidate path for mmdeploy_python * move "enable CUDA" to cmake config for demo * refine demo cmake * add comment * fix ubuntu build * revert docs/en/build.md * fix C API * fix lint * Windows build doc (#3) * check in docs related to mmdeploy build on windows * update build guide on windows platform * update build guide on windows platform * make path of thirdparty libraries consistent * make path consistency * correct build command for custom ops * correct build command for sdk * update sdk build instructions * update doc * correct build command * fix lint * correct build command and fix lint Co-authored-by: lvhan <lvhan@pjlab.org> * trailing whitespace (#4) * minor fix * fix sr sdk model * fix type deduction * fix cudaFree after driver shutting down * update ppl.cv installation warning (#5) * fix device allocator threshold & fix lint * update doc (#6) * update ppl.cv installation warning * missing 'git clone' Co-authored-by: chenxin <chenxin2@sensetime.com> Co-authored-by: zhangli <zhangli@sensetime.com> Co-authored-by: lvhan028 <lvhan_028@163.com> Co-authored-by: lvhan <lvhan@pjlab.org>
2022-02-24 20:08:44 +08:00
class MMDEPLOY_API ModelRegistry {
Merge sdk (#251) * check in cmake * move backend_ops to csrc/backend_ops * check in preprocess, model, some codebase and their c-apis * check in CMakeLists.txt * check in parts of test_csrc * commit everything else * add readme * update core's BUILD_INTERFACE directory * skip codespell on third_party * update trt_net and ort_net's CMakeLists * ignore clion's build directory * check in pybind11 * add onnx.proto. Remove MMDeploy's dependency on ncnn's source code * export MMDeployTargets only when MMDEPLOY_BUILD_SDK is ON * remove useless message * target include directory is wrong * change target name from mmdeploy_ppl_net to mmdeploy_pplnn_net * skip install directory * update project's cmake * remove useless code * set CMAKE_BUILD_TYPE to Release by force if it isn't set by user * update custom ops CMakeLists * pass object target's source lists * fix lint end-of-file * fix lint: trailing whitespace * fix codespell hook * remove bicubic_interpolate to csrc/backend_ops/ * set MMDEPLOY_BUILD_SDK OFF * change custom ops build command * add spdlog installation command * update docs on how to checkout pybind11 * move bicubic_interpolate to backend_ops/tensorrt directory * remove useless code * correct cmake * fix typo * fix typo * fix install directory * correct sdk's readme * set cub dir when cuda version < 11.0 * change directory where clang-format will apply to * fix build command * add .clang-format * change clang-format style from google to file * reformat csrc/backend_ops * format sdk's code * turn off clang-format for some files * add -Xcompiler=-fno-gnu-unique * fix trt topk initialize * check in config for sdk demo * update cmake script and csrc's readme * correct config's path * add cuda include directory, otherwise compile failed in case of tensorrt8.2 * clang-format onnx2ncnn.cpp Co-authored-by: zhangli <lzhang329@gmail.com> Co-authored-by: grimoire <yaoqian@sensetime.com>
2021-12-07 10:57:55 +08:00
public:
using Creator = std::function<std::unique_ptr<ModelImpl>()>;
struct Entry {
std::string name;
Creator creator;
};
/**
* @brief Return global instance of `ModelRegistry`
*/
Support Windows (#106) * minor changes * support windows * fix GCC build * fix lint * reformat * fix Windows build * fix GCC build * search backend ops for onnxruntime * fix lint * fix lint * code clean-up * code clean-up * fix clang build * fix trt support * fix cmake for ncnn * fix cmake for openvino * fix SDK Python API * handle ops for other backends (ncnn, trt) * handle SDK Python API library location * robustify linkage * fix cuda * minor fix for openvino & ncnn * use CMAKE_CUDA_ARCHITECTURES if set * fix cuda preprocessor * fix misc * fix pplnn & pplcv, drop support for pplcv<0.6.0 * robustify cmake * update build.md (#2) * build dynamic modules as module library & fix demo (partially) * fix candidate path for mmdeploy_python * move "enable CUDA" to cmake config for demo * refine demo cmake * add comment * fix ubuntu build * revert docs/en/build.md * fix C API * fix lint * Windows build doc (#3) * check in docs related to mmdeploy build on windows * update build guide on windows platform * update build guide on windows platform * make path of thirdparty libraries consistent * make path consistency * correct build command for custom ops * correct build command for sdk * update sdk build instructions * update doc * correct build command * fix lint * correct build command and fix lint Co-authored-by: lvhan <lvhan@pjlab.org> * trailing whitespace (#4) * minor fix * fix sr sdk model * fix type deduction * fix cudaFree after driver shutting down * update ppl.cv installation warning (#5) * fix device allocator threshold & fix lint * update doc (#6) * update ppl.cv installation warning * missing 'git clone' Co-authored-by: chenxin <chenxin2@sensetime.com> Co-authored-by: zhangli <zhangli@sensetime.com> Co-authored-by: lvhan028 <lvhan_028@163.com> Co-authored-by: lvhan <lvhan@pjlab.org>
2022-02-24 20:08:44 +08:00
static ModelRegistry& Get();
Merge sdk (#251) * check in cmake * move backend_ops to csrc/backend_ops * check in preprocess, model, some codebase and their c-apis * check in CMakeLists.txt * check in parts of test_csrc * commit everything else * add readme * update core's BUILD_INTERFACE directory * skip codespell on third_party * update trt_net and ort_net's CMakeLists * ignore clion's build directory * check in pybind11 * add onnx.proto. Remove MMDeploy's dependency on ncnn's source code * export MMDeployTargets only when MMDEPLOY_BUILD_SDK is ON * remove useless message * target include directory is wrong * change target name from mmdeploy_ppl_net to mmdeploy_pplnn_net * skip install directory * update project's cmake * remove useless code * set CMAKE_BUILD_TYPE to Release by force if it isn't set by user * update custom ops CMakeLists * pass object target's source lists * fix lint end-of-file * fix lint: trailing whitespace * fix codespell hook * remove bicubic_interpolate to csrc/backend_ops/ * set MMDEPLOY_BUILD_SDK OFF * change custom ops build command * add spdlog installation command * update docs on how to checkout pybind11 * move bicubic_interpolate to backend_ops/tensorrt directory * remove useless code * correct cmake * fix typo * fix typo * fix install directory * correct sdk's readme * set cub dir when cuda version < 11.0 * change directory where clang-format will apply to * fix build command * add .clang-format * change clang-format style from google to file * reformat csrc/backend_ops * format sdk's code * turn off clang-format for some files * add -Xcompiler=-fno-gnu-unique * fix trt topk initialize * check in config for sdk demo * update cmake script and csrc's readme * correct config's path * add cuda include directory, otherwise compile failed in case of tensorrt8.2 * clang-format onnx2ncnn.cpp Co-authored-by: zhangli <lzhang329@gmail.com> Co-authored-by: grimoire <yaoqian@sensetime.com>
2021-12-07 10:57:55 +08:00
/**
* @brief Register an sdk model format denoted by an specified `ModelImpl`
* @param name sdk model implementor's name
* @param creator method to create an sdk model implementor
* @return Status of registering result
*/
Result<void> Register(const std::string& name, Creator creator);
/**
* @brief Return the registered sdk model implementors
*/
const std::vector<Entry>& ListEntries() const { return entries_; }
private:
ModelRegistry() = default;
private:
std::vector<Entry> entries_;
};
} // namespace mmdeploy
#endif // !CORE_SDK_MODEL_H