2.3M的图片每次识别必定内存溢出

[blandDrift]

问题描述 / Problem Description

此图片在ocr时会直接内存溢出

-------------------------------------- 
C++ Traceback (most recent call last): 
-------------------------------------- 
paddle::AnalysisPredictor::ZeroCopyRun() 
paddle::framework::NaiveExecutor::Run() 
paddle::framework::OperatorBase::Run(paddle::framework::Scope const&, phi::Place const&) 
paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, phi::Place const&) const 
paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, phi::Place const&, paddle::framework::RuntimeContext*) const 
void phi::HardSwishKernel<float, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor*) 
void phi::ActivationGPUImpl<float, phi::GPUContext, phi::funcs::CudaHardSwishFunctor<float> >(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor*, phi::funcs::CudaHardSwishFunctor<float> const&) 
float* phi::DeviceContext::Alloc<float>(phi::TensorBase*, unsigned long, bool) const 
phi::DeviceContext::Impl::Alloc(phi::TensorBase*, phi::Place const&, phi::DataType, unsigned long, bool, bool) const 
phi::DenseTensor::AllocateFrom(phi::Allocator*, phi::DataType, unsigned long, bool) 
paddle::memory::allocation::Allocator::Allocate(unsigned long) 
paddle::memory::allocation::StatAllocator::AllocateImpl(unsigned long) 
paddle::memory::allocation::Allocator::Allocate(unsigned long) 
paddle::memory::allocation::Allocator::Allocate(unsigned long) 
paddle::memory::allocation::Allocator::Allocate(unsigned long) 
paddle::memory::allocation::Allocator::Allocate(unsigned long) 
paddle::memory::allocation::CUDAAllocator::AllocateImpl(unsigned long) 
std::string phi::enforce::GetCompleteTraceBackString<std::string >(std::string&&, char const*, int) 
phi::enforce::GetCurrentTraceBackString[abi:cxx11](bool)
---------------------- Error Message Summary: ---------------------- 
ResourceExhaustedError: Out of memory error on GPU 0. Cannot allocate 2.089966GB memory on GPU 0, 7.469543GB memory has been allocated and available memory is only 362.875000MB. 
Please check whether there is any other process using GPU 0. 
1. If yes, please stop them, or start PaddlePaddle on another GPU. 
2. If no, please decrease the batch size of your model. (at ../paddle/fluid/memory/allocation/cuda_allocator.cc:86)

运行环境 / Runtime Environment

Red Hat Enterprise Linux 8.10
python3.8.13
rapidocr-paddle 1.3.27
nvidia A10 8G

复现代码 / Reproduction Code

import cv2

from rapidocr_paddle import RapidOCR

# 注意这里的参数
engine = RapidOCR(det_use_cuda=True, cls_use_cuda=True, rec_use_cuda=True)

image_path = "tests/1.png"
result, elapse_list = engine(image_path)

可能解决方案 / Possible solutions

重启

[SWHL]

这个不是bug。
rapidocr是动态输入的，输入图像尺寸越大，机器会相应申请内存越大，申请完之后会释放掉。
所以解决方案有两个：
方案一：输入前根据需求裁剪或resize图像，再送入OCR中
方案二：增加机器内存

[lazwa34]

使用gpu推理，正常推理会自动释放显存。但是在批量任务中，某张图片尺寸太大导致显存溢出后并不会自动释放，后续任务都会报错 Out of memory。

--------------------------------------
C++ Traceback (most recent call last):
--------------------------------------
Not support stack backtrace yet.

----------------------
Error Message Summary:
----------------------
ResourceExhaustedError:

Out of memory error on GPU 0. Cannot allocate 266.406250MB memory on GPU 0, 1.999756GB memory has been allocated and available memory is only 0.000000B.

Please check whether there is any other process using GPU 0.
1. If yes, please stop them, or start PaddlePaddle on another GPU.
2. If no, please decrease the batch size of your model.
 (at ..\paddle\fluid\memory\allocation\cuda_allocator.cc:86)

[SWHL]

嗯嗯，这个目前推测为框架问题，暂时没找到好的解决方案