Custom trained rknn

General
1

i trained a custom model to pt. then converted to onnx and then rknn as given in GitHub - vicharak-in/Axon-NPU-Guide: This repository contains guide on how to setup toolkits to use NPU present on Axon for running various CNN models. the output is

Screenshot 2025-03-25 at 08-57-34 safety.rknn
Screenshot 2025-03-25 at 08-57-34 safety.rknn501×931 8.27 KB

i am trying to post process the output but getting the error " grid_h, grid_w = position.shape[2:4]
ValueError: not enough values to unpack (expected 2, got 1)
".

what am i doing wrong?

2

position.shape would return the shape of position which is tuple representing dimensions of position. your code requires position to be at least 4 dimensional, maybe it is of < 4 dimensional

3

position is of 3 dimension, so position.shape[2:4] is returning only value at position.shape[2], so it is returning only 1 value, but in left hand side you were expecting 2 values, so it is giving error