RKNNLITE does not accept float32 normalized images

Hello guys,

can someone help me with this

when you run inference with
rknn_lite.inference(inputs=[image])
rknn_lite expects image to be uint8 (0…255 values) and it internaly normalizes it to float32 (0…1) values …

but what if i am not working with images but some other input, that has normalized float32 input ( 0.0 … 1.0) values … (aka is already normalized by divinding it / 255.0).

I tried to put to input normalized image , and inference using rknn_lite is just wrong, even tho my network has as input float32 image…

ill answer it myself … its inputs_pass_through parameter in rknn_lite.inference

1 Like