Question/Issue: Currently edgeimpulse is supporting custom models for detectors like Yolo/SSD with BYOM. We have an architecture resembling Centernet and has been hand customized to fit under 500KB of activation memory and weight parameter under 1MB. We want to run this customized detector in OpenMV hardware. Is there a way to run this model? If so, how would you recommend partitioning the model for tflite and post processing involved for detections?
Project ID: Object Detector
Context/Use case: Object Detection with bounding box
Given the complexity and customization required for deploying your custom Centernet detector on OpenMV hardware, I would recommend considering doing our Enterprise tier trial. The Enterprise tier offers advanced features and tailored support to meet the specific needs of projects like yours.
I will try to get answers to your queries, and answer them in this thread but let me also link you in via email with our dedicated APAC sales team so you can be aware of our dedicated support options.