自行编译的fastdeploy无法使用gpu进行推理

[tammy-na]

环境

• 【编译命令】自行编译的FastDeploy，（参数命令）：

cd FastDeploy/python
export ENABLE_ORT_BACKEND=ON
export ENABLE_PADDLE_BACKEND=ON
export ENABLE_OPENVINO_BACKEND=ON
export ENABLE_VISION=ON
export ENABLE_TEXT=ON
export ENABLE_TRT_BACKEND=ON
export WITH_GPU=ON
export TRT_DIRECTORY=/usr/local/TensorRT-8.5.2.2
export CUDA_DIRECTORY=/usr/local/cuda-11.7
export OPENCV_DIRECTORY=/usr/lib/x86_64-linux-gnu/cmake/opencv4

python setup.py build
python setup.py bdist_wheel

• 【系统平台】: Linux x64(Ubuntu 20.04.1)
• 【硬件】： Nvidia GPU 3060， CUDA 11.7 CUDNN 8.6.0
• 【编译语言】：Python3.8

问题日志及出现问题的操作流程

cd FastDeploy/examples/vision/ocr/PP-OCR/cpu-gpu/python
python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv3_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device gpu --backend paddle
[INFO] fastdeploy/runtime/backends/paddle/paddle_backend.cc(29)::BuildOption Will inference_precision float32
[INFO] fastdeploy/runtime/runtime.cc(286)::CreatePaddleBackend Runtime initialized with Backend::PDINFER in Device::GPU.
[INFO] fastdeploy/runtime/backends/paddle/paddle_backend.cc(29)::BuildOption Will inference_precision float32
[INFO] fastdeploy/runtime/runtime.cc(286)::CreatePaddleBackend Runtime initialized with Backend::PDINFER in Device::GPU.
[INFO] fastdeploy/runtime/backends/paddle/paddle_backend.cc(29)::BuildOption Will inference_precision float32
[INFO] fastdeploy/runtime/runtime.cc(286)::CreatePaddleBackend Runtime initialized with Backend::PDINFER in Device::GPU.
free(): invalid size
已放弃 (核心已转储)
问题总结：当--device gpu 时会中断推理，当--device cpu时不存在问题：
python infer.py --det_model ch_PP-OCRv3_det_infer --cls_model ch_ppocr_mobile_v2.0_cls_infer --rec_model ch_PP-OCRv3_rec_infer --rec_label_file ppocr_keys_v1.txt --image 12.jpg --device cpu --backend paddle
[INFO] fastdeploy/runtime/runtime.cc(286)::CreatePaddleBackend Runtime initialized with Backend::PDINFER in Device::CPU.
[INFO] fastdeploy/runtime/runtime.cc(286)::CreatePaddleBackend Runtime initialized with Backend::PDINFER in Device::CPU.
[INFO] fastdeploy/runtime/runtime.cc(286)::CreatePaddleBackend Runtime initialized with Backend::PDINFER in Device::CPU.
det boxes: [[42,413],[483,391],[484,428],[43,450]]rec text: 上海斯格威铂尔大酒店 rec score:0.980085 cls label: 0 cls score: 1.000000
det boxes: [[187,456],[399,448],[400,480],[188,488]]rec text: 打浦路15号 rec score:0.964993 cls label: 0 cls score: 1.000000
det boxes: [[23,507],[513,488],[515,529],[24,548]]rec text: 绿洲仕格维花园公寓 rec score:0.993727 cls label: 0 cls score: 1.000000
det boxes: [[74,553],[427,542],[428,571],[75,582]]rec text: 打浦路252935号 rec score:0.947723 cls label: 0 cls score: 1.000000

当我把自行编译的fastdeploy卸载之后安装：pip install fastdeploy-gpu-python==1.0.7 -f https://www.paddlepaddle.org.cn/whl/fastdeploy.html，这时gpu和cpu都可以推理，我为什么非要重新编译fastdeploy是因为我想要添加如下内容到fastdeploy/vision/vision_pybind.cc文件中。
.def_readwrite("table_html", &vision::OCRResult::table_structure)
.def_readwrite("table_boxes", &vision::OCRResult::table_boxes)

[Jiang-Jia-Jun]

Linux吗，你可以试下使用Paddle官方的镜像来编译