YOLOV8n conversion to tflite runs only on the CPU but has segmentation errors trying to run on the GPU

ErezI

Hi,
We tried to convert YOLOV8n detection to tflite on VOXL2 according to your instruction
(https://docs.modalai.com/voxl-tflite-server/).

After changing also the new tflite to the same existing YOLOV5 file name
(/usr/bin/dnn/yolov5_float16_quant.tflite)
we succeed in running it with CPU delegate
(/etc/modalai/voxl-tflite-server.conf),
but not on the GPU.

1.Is it possible to get the full yolov5 to tflite conversation
you used in order to get a functional tflite file running on the GPU
and also the relevant configuration and parameters used for the already existing YOLOV5's tflite file
(conf, iou, opset, imgsz, max_det, half, int8, dnn, etc.).

2.Did you try to convert & use updated YOLO version for the VOXL2 GPU?

Thanks,

Erez

Moderator

@ErezI this thread is likely useful https://forum.modalai.com/topic/2997/can-anyone-recommend-a-tflite-colab-notebook-for-voxl2-training/28

@ErezI

After following that thread let me know what questions you have, I'm happy to help!!

Thomas Patton