apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-11-02 20:54:03 +08:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

[Bug Fix] add ocr new feature and fix codestyle (#764)

Browse Source 
* fix ocr bug and add new feature

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* add property

* add test

* fix code style

* fix bug

* fix bug

* fix bug

* fix port

* fix ocr

* fix_ocr

* fix ocr

* fix ocr

* fix ocr

* fix ocr

* Update paddle2onnx.cmake

* Update paddle2onnx.cmake

* Update paddle2onnx.cmake

Co-authored-by: Jason <jiangjiajun@baidu.com>
Co-authored-by: Jason <928090362@qq.com>
This commit is contained in:
Thomas Young
2022-12-07 19:31:54 +08:00
committed by GitHub
parent e6af8f2334
commit 5df62485c3
33 changed files with 1222 additions and 376 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
fastdeploy/vision/ocr/ppocr/rec_preprocessor.cc
View File

@@ -21,69 +21,75 @@ namespace fastdeploy {
namespace vision {
namespace ocr {
RecognizerPreprocessor::RecognizerPreprocessor() {
initialized_ = true;
}
void OcrRecognizerResizeImage(FDMat* mat, float max_wh_ratio,
const std::vector<int>& rec_image_shape) {
int imgC, imgH, imgW;
imgC = rec_image_shape[0];
imgH = rec_image_shape[1];
imgW = rec_image_shape[2];
int img_c, img_h, img_w;
img_c = rec_image_shape[0];
img_h = rec_image_shape[1];
img_w = rec_image_shape[2];
imgW = int(imgH * max_wh_ratio);
img_w = int(img_h * max_wh_ratio);
float ratio = float(mat->Width()) / float(mat->Height());
int resize_w;
if (ceilf(imgH * ratio) > imgW) {
resize_w = imgW;
if (ceilf(img_h * ratio) > img_w) {
resize_w = img_w;
}else{
resize_w = int(ceilf(imgH * ratio));
resize_w = int(ceilf(img_h * ratio));
}
Resize::Run(mat, resize_w, imgH);
Resize::Run(mat, resize_w, img_h);
std::vector<float> value = {0, 0, 0};
Pad::Run(mat, 0, 0, 0, int(imgW - mat->Width()), value);
Pad::Run(mat, 0, 0, 0, int(img_w - mat->Width()), value);
}
bool RecognizerPreprocessor::Run(std::vector<FDMat>* images, std::vector<FDTensor>* outputs) {
if (!initialized_) {
FDERROR << "The preprocessor is not initialized." << std::endl;
return false;
}
if (images->size() == 0) {
FDERROR << "The size of input images should be greater than 0." << std::endl;
return Run(images, outputs, 0, images->size(), {});
}
bool RecognizerPreprocessor::Run(std::vector<FDMat>* images, std::vector<FDTensor>* outputs,
size_t start_index, size_t end_index, const std::vector<int>& indices) {
if (images->size() == 0 || end_index <= start_index || end_index > images->size()) {
FDERROR << "images->size() or index error. Correct is: 0 <= start_index < end_index <= images->size()" << std::endl;
return false;
}
int imgH = rec_image_shape_[1];
int imgW = rec_image_shape_[2];
float max_wh_ratio = imgW * 1.0 / imgH;
int img_h = rec_image_shape_[1];
int img_w = rec_image_shape_[2];
float max_wh_ratio = img_w * 1.0 / img_h;
float ori_wh_ratio;
for (size_t i = 0; i < images->size(); ++i) {
FDMat* mat = &(images->at(i));
for (size_t i = start_index; i < end_index; ++i) {
size_t real_index = i;
if (indices.size() != 0) {
real_index = indices[i];
}
FDMat* mat = &(images->at(real_index));
ori_wh_ratio = mat->Width() * 1.0 / mat->Height();
max_wh_ratio = std::max(max_wh_ratio, ori_wh_ratio);
}
for (size_t i = 0; i < images->size(); ++i) {
FDMat* mat = &(images->at(i));
for (size_t i = start_index; i < end_index; ++i) {
size_t real_index = i;
if (indices.size() != 0) {
real_index = indices[i];
}
FDMat* mat = &(images->at(real_index));
OcrRecognizerResizeImage(mat, max_wh_ratio, rec_image_shape_);
NormalizeAndPermute::Run(mat, mean_, scale_, is_scale_);
/*
Normalize::Run(mat, mean_, scale_, is_scale_);
HWC2CHW::Run(mat);
Cast::Run(mat, "float");
*/
}
// Only have 1 output Tensor.
outputs->resize(1);
size_t tensor_size = end_index-start_index;
// Concat all the preprocessed data to a batch tensor
std::vector<FDTensor> tensors(images->size());
for (size_t i = 0; i < images->size(); ++i) {
(*images)[i].ShareWithTensor(&(tensors[i]));
std::vector<FDTensor> tensors(tensor_size);
for (size_t i = 0; i < tensor_size; ++i) {
size_t real_index = i + start_index;
if (indices.size() != 0) {
real_index = indices[i + start_index];
}
(*images)[real_index].ShareWithTensor(&(tensors[i]));
tensors[i].ExpandDim(0);
}
if (tensors.size() == 1) {