Custom YOLOv8 model hard faults on N6 when loaded via ml.Model()

OpenMV Boards
1

Device: OpenMV N6, firmware v4.8.1, MicroPython v1.26.0-77

Issue: When I add a custom INT8 quantized YOLOv8 tflite model to ROMFS
via the IDE editor, the IDE automatically converts it to a bin file
(3,236,416 bytes) using ST EdgeAI. But calling ml.Model() on this file
causes a hard fault and board reset.

The official yolov8n_192.tflite (3,283,296 bytes, original tflite format,
not converted) loads fine with ml.Model().

My model: YOLOv8n, 10 classes, input 224x224, INT8 full integer quantized.

Question: How do I load a custom model that has been converted to bin
format by the IDE? Or how do I prevent the IDE from converting and keep
the original tflite in ROMFS?

2

You need to update the the latest N6 firmware.

Please do: Tools->Install the Latest Development Release. Then the new model will work.

3

Thank you so much!My custom YOLOv8 model now loads successfully with ml.Model().