mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-10-06 09:07:10 +08:00
c6480de736
* bind success * bind success fix * FDMat pybind, ResizeByShort pybind * FDMat pybind, ResizeByShort pybind, remove initialized_ * override BindProcessorManager::Run in python is available * PyProcessorManager done * vision_pybind fix * manager.py fix * add tutorials * remove Apply() bind * remove Apply() bind and fix * fix reviewed problem * fix reviewed problem * fix reviewed problem readme * fix reviewed problem readme etc * apply return outputs * nits * update readme * fix FDMatbatch * add op pybind: CenterCrop, Pad * add op overload for pass FDMatBatch --------- Co-authored-by: Wang Xinyu <shaywxy@gmail.com>
82 lines
2.4 KiB
C++
82 lines
2.4 KiB
C++
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
#pragma once
|
|
#include "fastdeploy/vision/common/processors/mat.h"
|
|
|
|
#ifdef WITH_GPU
|
|
#include <cuda_runtime_api.h>
|
|
#endif
|
|
|
|
namespace fastdeploy {
|
|
namespace vision {
|
|
|
|
enum FDMatBatchLayout { NHWC, NCHW };
|
|
|
|
struct FASTDEPLOY_DECL FDMatBatch {
|
|
FDMatBatch() = default;
|
|
|
|
// MatBatch is intialized with a list of mats,
|
|
// the data is stored in the mats separately.
|
|
// Call Tensor() function to get a batched 4-dimension tensor.
|
|
explicit FDMatBatch(std::vector<FDMat>* _mats) {
|
|
mats = _mats;
|
|
layout = FDMatBatchLayout::NHWC;
|
|
mat_type = ProcLib::OPENCV;
|
|
}
|
|
|
|
// Get the batched 4-dimension tensor.
|
|
FDTensor* Tensor();
|
|
|
|
void SetTensor(FDTensor* tensor);
|
|
|
|
private:
|
|
#ifdef WITH_GPU
|
|
cudaStream_t stream = nullptr;
|
|
#endif
|
|
std::shared_ptr<FDTensor> fd_tensor = std::make_shared<FDTensor>();
|
|
|
|
public:
|
|
// When using CV-CUDA/CUDA, please set input/output cache,
|
|
// refer to manager.cc
|
|
FDTensor* input_cache;
|
|
FDTensor* output_cache;
|
|
#ifdef WITH_GPU
|
|
cudaStream_t Stream() const { return stream; }
|
|
void SetStream(cudaStream_t s);
|
|
#endif
|
|
|
|
std::vector<FDMat>* mats = nullptr;
|
|
|
|
// Used by pybind, since python cannot pass list as pointer or reference
|
|
std::vector<FDMat> mats_holder;
|
|
|
|
ProcLib mat_type = ProcLib::OPENCV;
|
|
FDMatBatchLayout layout = FDMatBatchLayout::NHWC;
|
|
Device device = Device::CPU;
|
|
ProcLib proc_lib = ProcLib::DEFAULT;
|
|
|
|
// False: the data is stored in the mats separately
|
|
// True: the data is stored in the fd_tensor continuously in 4 dimensions
|
|
bool has_batched_tensor = false;
|
|
};
|
|
|
|
// Create a batched input tensor on GPU and save into input_cache.
|
|
// If the MatBatch is on GPU, return the Tensor() directly.
|
|
// If the MatBatch is on CPU, then copy the CPU tensors to GPU and get a GPU
|
|
// batched input tensor.
|
|
FDTensor* CreateCachedGpuInputTensor(FDMatBatch* mat_batch);
|
|
|
|
} // namespace vision
|
|
} // namespace fastdeploy
|