We’ve run into a problem with converting our model from Torch to RKNN. It converts fine but the converted model outputs significantly different results to the Torch model.
On the same system we’ve tried converting the demo models and they work fine. As far as we’re aware all of our layers are supported.
We are converting it with the most basic options, but have also tried as many other available options as well and can’t find a good combination -
rk.config(batch_size=1, optimize=0) rk.build(do_quantization=False)
Are you able to offer any guidance for where we should go for support or how we can troubleshoot this ourselves?