I bought luckfox pico ultra W, burned Buildroot, I can use yolov5 normally according to the routine according to the wiki's operation RKMPI instance user guide, I try to train the model by myself, and convert it to onnx, convert it to rknn again, I replace the rknn file I got luckfox_pico_rkmpi_example/example/luckfox_pico_ rtsp_yolov5 yolov5.rknn in the /model$ directory, run ./build.sh, upload the file generated under install to the development board, and the error message "Segmentation fault (core dumped)" is displayed.
The address of the warehouse where I trained yolov5 was https://github.com/airockchip/yolov5
There was a problem deploying your own yolov5 model
-
- Attachments
-
Hello, there are many possibilities for segfault when running a self-trained model, and there are problems and solutions that are usually to encounter in the RKNN runtime in https://wiki.luckfox.com/Luckfox-Pico/Luckfox-Pico-FAQ/, you can check them.
It is recommended to run the model first using the luckfox_pico_rknn (https://github.com/LuckfoxTECH/luckfox_ ... nn_example) routine, your program occurs during the process of RKMPI initialization, and there is a possibility that RKMPI is misconfigured
It is recommended to run the model first using the luckfox_pico_rknn (https://github.com/LuckfoxTECH/luckfox_ ... nn_example) routine, your program occurs during the process of RKMPI initialization, and there is a possibility that RKMPI is misconfigured
Thank you, my problem is solved, according to the solution in FAQ, I adjusted Modify `model/coco_80_labels_list.txt` with the names of the objects recognized by the model (one object per line) and adjust the `OBJ_CLASS_NUM` definition in `include/postprocess.h` to match the number of recognized object categories.