My onnx yolo model keeps getting bloated on edge impulse from 11.8 MB when exported to about 75 MB

Report bugs
#1

Question/Issue: I have loaded my yolo onnx model by using the BYOM feature of edge impulse. I also have uploaded the represenative feature file to make use of quantizations. My XIAO ESP S3 32 chips can only store about 8MB directly and an additional 8MB with flash giving me space for 16MB. My model when downloaded from my code and loaded intitially was 11.8 MB. However, the C++ library I have downloaded results in a tflite model that is 73MB! Am I missing any step that’s causing this bloat?

**Project ID:**339277

Context/Use case: Glasses using YOLO for object detection and then speech conversion for visually impaired people

#2

Hi @abhiramt4009

The increase in size from the ONNX model to the TensorFlow Lite (tflite) model might be due to the conversion process, where certain optimizations or quantizations are not applied, or additional metadata is included. Did you enable eon compiler and 8 bit quantisation?

Best

Eoin