Luckfox Forums

Hello everyone I bought a Luckfox pico ultra w (rv1106). I made my dataset for 2 classes, using the code from the official repository- yolov5_pico_rtsp_demo. I can't adapt the open source code to my task in any way. Maybe someone here has their own ideas that they can share? It seems to me that the whole problem is in postprocessing.


P>S

Well, I was able to make normal weights after all - thanks to this fork. And how to adapt the code for 2 classes, gentlemen?

Hello! If the structure of the final output layer is (1, 22743, 7), it indicates that the model you are using has two categories.
Please modify the OBJ_CLASS_NUM in $luckfox_pico_rkmpi_example/include/rknn/postprocess.h to 2.
Also, modify example/luckfox_pico_rtsp_yolov5/model/coco_80_labels_list.txt to only retain the names of the categories that you can recognize.