I’m working on deploying a custom-trained Anomalib FastFlow model on a Hailo processor (Raspberry Pi 5) . I successfully exported my model from ‘.ckpt’ to ‘.onnx’ and then parsed it to ‘.har’. However, when trying to quantize the model before compiling to .hef, I ran into an issue.
The Hailo model optimization step fails during the statistics collection phase, throwing the following error:
indices[0] = [-2147483648] does not index into shape [1000]
It looks like TensorFlow is encountering an out-of-range index issue when using ‘TensorScatterAd’.