craig parsey’s Post

View profile for craig parsey, graphic

Heavy duty fitter and augmented intelligence embodied ai augmented reality

so working with a blank yolo model quantized to int8 was a bad idea.The edge hardware I'm using meant I took some shortcuts as training was taking to long.Anyway I ended up downloading a pre trained model and running conversion to int8 tflite model as I usually just add a class I must say tensorflow just works now on Windows every time I've had to download the tensorflow library I've always resorted to docker. I was going to run onnx but I've had issues in the path going onnx then int8 with models but man tensorflow names are weird so it makes a few flavours you'd say some partial int8 one full int8 the model called yoloint8.tflite isn't the full int8 no no no it's the yolo11_full_interger_quant.tflite. That makes perfect sense why would I stupidly think the model called int8 would be the actual int8 model. But now tensorflow is back again just working on Windows (with some downloaded dependencies onnx,c++) I think it's back to being usefull again for me as it can be packaged and shipped without sending someone to location to get running.

To view or add a comment, sign in

Explore topics