Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Modify PPMatting backend and docs #182

Merged
merged 82 commits into from
Sep 7, 2022
Merged

Conversation

ziqi-jin
Copy link
Contributor

@ziqi-jin ziqi-jin commented Sep 1, 2022

No description provided.

@@ -50,39 +50,6 @@ void CpuInfer(const std::string& model_dir, const std::string& image_file,
<< std::endl;
}

void GpuInfer(const std::string& model_dir, const std::string& image_file,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里保留GpuInfer,但是设置一下backend为paddle

option.UsePaddleBackend()

@@ -131,8 +98,6 @@ int main(int argc, char* argv[]) {
}
if (std::atoi(argv[4]) == 0) {
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

同上,保留gpu推理案例

@@ -19,8 +19,6 @@ wget https://bj.bcebos.com/paddlehub/fastdeploy/matting_input.jpg
wget https://bj.bcebos.com/paddlehub/fastdeploy/matting_bgr.jpg
# CPU推理
python infer.py --model PP-Matting-512 --image matting_input.jpg --bg matting_bgr.jpg --device cpu
# GPU推理 (TODO: ORT-GPU 推理会报错)
python infer.py --model PP-Matting-512 --image matting_input.jpg --bg matting_bgr.jpg --device gpu
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

同上

@@ -27,8 +27,6 @@ wget https://bj.bcebos.com/paddlehub/fastdeploy/matting_bgr.jpg

# CPU推理
./infer_demo PP-Matting-512 matting_input.jpg matting_bgr.jpg 0
# GPU推理 (TODO: ORT-GPU 推理会报错)
./infer_demo PP-Matting-512 matting_input.jpg matting_bgr.jpg 1
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里保留GpuInfer,但是设置一下backend为paddle

@@ -92,6 +96,22 @@ bool PPMatting::BuildPreprocessPipelineFromConfig() {
std = op["std"].as<std::vector<float>>();
}
processors_.push_back(std::make_shared<Normalize>(mean, std));
} else if (op["type"].as<std::string>() == "ResizeByLong") {
int target_size = 512;
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

为啥要给默认值

std::cout << "If LimintShort in yaml file, you may transfer PPMatting "
"model by yourself, please make sure your input image's "
"width==hight and not smaller than "
<< max_short << std::endl;
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

日志输出改为FDINFO

LimitShort, height 单词拼错了

改为Detected LimitShort processing step in yaml file, if the model is exported from PaddleSeg, please make sure the input of your model is fixed with a square shape, and equal to " << max_short << "." << std::endl;

} else if (op["type"].as<std::string>() == "Pad") {
// size: (w, h)
auto size = op["size"].as<std::vector<int>>();
std::vector<float> value = {114, 114, 114};
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

为啥是114, 114的这个默认值是PaddleSeg里面默认的吗?

if (processors_[i]->Name().compare("LimitShort") == 0) {
int input_h = static_cast<int>(mat->Height());
int input_w = static_cast<int>(mat->Width());
FDASSERT(input_h == input_w,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里用FDASSERT,用户输入的如果不是方形的,是不是会直接报错了?我看到这个input_h/w,是从输入的mat获取的,并不是从模型输入的shape获取的,但是报错提示的是"model"的input_shape必须是方形的

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

if (option.backend == Backend::PDINFER) {
    if (option.device == Device::CPU) {
         FDWARN << "" << std::endl;
        // some ops
    } 
}  else {
   FDWARN << "" << std::endl;
   // resize op
}

if (pad_to_size != im_info.end() && resize_by_long != im_info.end()) {
int resize_h = resize_by_long->second[0];
int resize_w = resize_by_long->second[1];
int pad_h = pad_to_size->second[0];
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里pad_h和pad_w获取后并没有用上,所以这两个值的具体作用是什么

"sure the input of your model is fixed with a square shape");
auto processor = dynamic_cast<LimitShort*>(processors_[i].get());
int max_short = processor->GetMaxShort();
FDASSERT(input_h >= max_short && input_w >= max_short,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

同上,这里获取的是输入的input_h和input_w,并不是模型的

if args.device.lower() == "gpu":
option.use_gpu()
if args.use_trt:
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个paddle gpu的判断,应该不用缩进一个层级,可以这样(gpu下,如果不指定trt,默认就用paddle推)

def build_option(args):
    option = fd.RuntimeOption()

    if args.device.lower() == "gpu":
        option.use_gpu()
        option.use_paddle_backend()

     if args.use_trt:
         option.use_trt_backend()
         option.set_trt_input_shape("img", [1, 3, 512, 512])
      
    return option

ProcLib lib = ProcLib::OPENCV_CPU);

private:
double GenerateScale(const int origin_w, const int origin_h);
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

naive 值传递 不需要 const限制

@jiangjiajun jiangjiajun merged commit 7e00c5f into PaddlePaddle:develop Sep 7, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants