apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-12-24 13:28:13 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[CVCUDA] PaddleDetection preprocessor support CV-CUDA (#1493)

Browse Source 
* ppdet preproc use manager

* pad_to_size chw opencv

* pad_to_size chw flycv

* fix pad_to_size flycv

* add warning message

* cvcuda convert cubic to linear, padToSize cvcuda

* stridedpad cvcuda

* fix flycv include

* fix flycv include

* fix flycv build

* cast cvcuda

* fix pybind

* fix normalize permute cuda

* base processor move funcs to cc

* Update pad_to_size.cc
This commit is contained in:
Wang Xinyu
2023-03-10 12:43:57 +08:00
committed by GitHub
parent 9ee2118e1f
commit cb7c8a07d4
23 changed files with 537 additions and 239 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
fastdeploy/vision/common/processors/stride_pad.cc
View File

@@ -114,6 +114,68 @@ bool StridePad::ImplByFlyCV(Mat* mat) {
}
#endif
#ifdef ENABLE_CVCUDA
bool StridePad::ImplByCvCuda(FDMat* mat) {
if (mat->layout != Layout::HWC) {
FDERROR << "StridePad: The input data must be Layout::HWC format!"
<< std::endl;
return false;
}
if (mat->Channels() > 4) {
FDERROR << "StridePad: Only support channels <= 4." << std::endl;
return false;
}
if (mat->Channels() != value_.size()) {
FDERROR
<< "StridePad: Require input channels equals to size of padding value, "
"but now channels = "
<< mat->Channels() << ", the size of padding values = " << value_.size()
<< "." << std::endl;
return false;
}
int origin_w = mat->Width();
int origin_h = mat->Height();
int pad_h = (mat->Height() / stride_) * stride_ +
(mat->Height() % stride_ != 0) * stride_ - mat->Height();
int pad_w = (mat->Width() / stride_) * stride_ +
(mat->Width() % stride_ != 0) * stride_ - mat->Width();
if (pad_h == 0 && pad_w == 0) {
return true;
}
float4 value;
if (value_.size() == 1) {
value = make_float4(value_[0], 0.0f, 0.0f, 0.0f);
} else if (value_.size() == 2) {
value = make_float4(value_[0], value_[1], 0.0f, 0.0f);
} else if (value_.size() == 3) {
value = make_float4(value_[0], value_[1], value_[2], 0.0f);
} else {
value = make_float4(value_[0], value_[1], value_[2], value_[3]);
}
// Prepare input tensor
FDTensor* src = CreateCachedGpuInputTensor(mat);
auto src_tensor = CreateCvCudaTensorWrapData(*src);
int height = mat->Height() + pad_h;
int width = mat->Width() + pad_w;
// Prepare output tensor
mat->output_cache->Resize({height, width, mat->Channels()}, mat->Type(),
"output_cache", Device::GPU);
auto dst_tensor = CreateCvCudaTensorWrapData(*(mat->output_cache));
cvcuda_pad_op_(mat->Stream(), src_tensor, dst_tensor, 0, 0,
NVCV_BORDER_CONSTANT, value);
mat->SetTensor(mat->output_cache);
mat->mat_type = ProcLib::CVCUDA;
return true;
}
#endif
bool StridePad::Run(Mat* mat, int stride, const std::vector<float>& value,
ProcLib lib) {
auto p = StridePad(stride, value);