C++ demo is provided in the cpp directory for reference, the details are as follows:
| No. | C++ demo | Description |
|---|---|---|
| 1 | yolov5_bmcv | Decoding with FFmpeg, preprocessing with BMCV, inference with BMRT |
| 2 | yolov5_sail | Decoding with SAIL, preprocessing with SAIL, inference with SAIL |
If you have installed a PCIe accelerator card (such as an SC series accelerator card) on the x86/arm platform, you can directly use it as a development environment and a runtime environment. You need to install libsophon, sophon-opencv and sophon-ffmpeg. For specific steps, please refer to Construction of Development and Runtime Environment of x86-PCIe Platform or Construction of Development and Runtime Environment of arm-PCIe Platform or Construction of Development and Runtime Environment of riscv-PCIe Platform.
If you use the SoC platform (such as SE, SM series edge devices), the corresponding libsophon, sophon-opencv and sophon-ffmpeg runtime packages have been pre-installed under /opt/sophon/ after resetting the operating system, and you can use it directly as runtime environment. Usually an x86 host is also required as a development environment for cross-compiling C++ programs.
C++ programs need to compile executable files before running.
Programs can be compiled directly on the PCIe platform:
cd cpp/yolov5_bmcv
mkdir build && cd build
cmake ..
make
cd ..After the compilation is complete, yolov5_bmcv.pcie will be generated in the yolov5_bmcv directory.
If you use sophon-sail, you should refer to sophon-sail install guide，configure sophon-sail for your sys，then:
cd cpp/yolov5_sail
mkdir build && cd build
cmake ..
make
cd ..After the compilation is complete, yolov5_sail.pcie will be generated in the yolov5_bmcv directory.
The cross-compilation program is usually on the x86 host. User need to use the SOPHON SDK to build a cross compilation environment on the x86 host, and package the header files and library files that the program depends on into the soc-sdk directory. For details, please refer to Cross-Compilation Environment Construction. This demo mainly depends on libsophon, sophon-opencv and sophon-ffmpeg runtime library packages.
After the cross compilation environment is set up, use the cross compilation toolchain to compile and generate executable files:
cd cpp/yolov5_bmcv
mkdir build && cd build
#Please modify the path of -DSDK according to the actual situation, here need to use the absolute path.
cmake -DTARGET_ARCH=soc -DSDK=/path_to_sdk/soc-sdk ..
makeAfter the compilation is complete, yolov5_bmcv.soc will be generated in the yolov5_bmcv directory.
If you use sophon-sail, you should refer to cross compile sophon-sail，configure sophon-sail for your soc，then:
cd cpp/yolov5_sail
mkdir build && cd build
#Please modify the path of -DSDK and -DSAIL according to the actual situation, here need to use the absolute path.
cmake -DTARGET_ARCH=soc -DSDK=/path_to_sdk/soc-sdk -DSAIL_PATH=/path_to_sail/sophon-sail/build_soc/sophon-sail ..
makeAfter the compilation is complete, yolov5_sail.soc will be generated in the yolov5_bmcv directory.
For the PCIe platform, user can directly reason and test on the PCIe platform. For the SoC platform, the executable file generated by cross compilation and the required model and test data need to be copied to the SoC platform for testing. The parameters and operation modes of the inference tests on the two platforms are the same, and the following will mainly introduce the PCIe mode.
The executable program has a set of parameters by default. Please pay attention to pass the parameters according to the actual situation. yolov5_bmcv.pcie and yolov5_sail.pcie has same parameters. The specific parameters are described as follows:
Usage: yolov5_bmcv.pcie [params]
--bmodel (value:../../models/BM1684/yolov5s_v6.1_3output_fp32_1b.bmodel)
bmodel file path
--classnames (value:../../datasets/coco.names)
class names file path
--conf_thresh (value:0.5)
confidence threshold for filter boxes
--dev_id (value:0)
TPU device id
--help (value:true)
print help information.
--input (value:../../datasets/test)
input path, images direction or video file path
--nms_thresh (value:0.5)
iou threshold for nms
--obj_thresh (value:0.5)
object score threshold for filter boxes
--use_cpu_opt (value:true)
accelerate cpu postprocessNote: The method of passing parameters in C++ is different from that in Python. The method of passing parameters in C++ needs to use the equal sign. For example
./yolov5_bmcv.pcie --bmodel=xxx. C++ demo use--use_cpu_opt=trueto activate cpu postprocess acceleration,use_cpu_optonly for model's outputs with 5 dimensions(normally 3 outputs model, if your model has more or less outputs, you should modify postprocess code by your self).
The image test demo is as follows. It supports testing the entire image folder. The model supports fp32bmodel and int8bmodel, supports BM1684 and BM1684X, and supports single batch size or multiple batch sizes. Users can run model testing by passing in the corresponding model path parameters.
./yolov5_bmcv.pcie --input=../../datasets/test --bmodel=../../models/BM1684/yolov5s_v6.1_3output_fp32_1b.bmodel --dev_id=0 --conf_thresh=0.5 --nms_thresh=0.5 --obj_thresh=0.5 --classnames=../../datasets/coco.names After the test, the predicted images will be saved in results/images, the predicted results will be saved in results/yolov5s_v6.1_3output_fp32_1b.bmodel_test_bmcv_cpp_result.json, and the prediction results, inference time and other information will be printed at the same time.
Note：
1.cpp example do not write on frame.
The video test demo is as follows, which supports testing of video streams. The model supports fp32bmodel and int8bmodel, supports BM1684 and BM1684X, and supports single batch size or multiple batch sizes. Users can run model testing by passing in the corresponding model path parameters.
./yolov5_bmcv.pcie --input=../../datasets/test_car_person_1080P.mp4 --bmodel=../../models/BM1684/yolov5s_v6.1_3output_fp32_1b.bmodel --dev_id=0 --conf_thresh=0.5 --nms_thresh=0.5 --obj_thresh=0.5 --classnames=../../datasets/coco.namesAfter the test, the prediction results will be drawn on the images and saved in results/images, and the prediction results, inference time and other information will be printed at the same time.
Note:
If this error is reported on SoC:
./yolov5_sail.soc: error while loading shared libraries: libsail.so: cannot open shared object file: No such file or directoryPlease do this:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/sophon/sophon-sail/lib