working on a CUSTOM RK3568 board and hitting an error on rknn_init. Using the ‘example’ from rknn_tookit2 for yolov5. Same code works on my custom RK3588 board. have ordered the Radxa 3A, but has not yet arrived, so asking this question in general for possible, well-known errors, or other tips to chase down the problem.
The converted, quantized yolov5 model is 10MB, free ram on the board id 1.7GB. I can load the model from file, but then calling rknn_init returns -4 - malloc error.
As noted, this exact same code works on RK 3588. Any tips? Tricks? Things to look at?
FYI - we build linux for this board with a bitbake/yocto project from Rockchip rk-linux. Not all of the source is available (notably the source for librknnrt.so which has the rknn_init code).