I am having issues with the fully integer quantized MiDaS model. The aim is to run it on coral edge tpu, however the results are inaccurate.
I tried to convert the model using OpenVINO mo.py script and the openvino2tensorflow toll. The results are inaccurate compared with the float32 model. I provided an image of the model's full_integer_quant.tflite output. enter image description here.