I am using Hailo 8 and the NUYV2 dataset for running depth estimation inference using the fast-depth pre compiled model which i downloaded from Model Explorer .
However after the inference the output tensor that i am getting is uint16
Which looks like this
Output Tensor shape: (1, 224, 224, 1)
Output Tensor: [[[[2741]
[2741]
[2862]
…
[2621]
[2570]
[2570]]
[[2741]
[2741]
[2862]
…
[2621]
[2570]
[2570]]
[[2784]
[2784]
[2896]
…
[2492]
[2604]
[2604]]
….
But i am not able to figure out how do i dequantise it to match the ground truth tensor which has floating values.
The ground truth tensor looks like
[[1.8842598 1.884209 1.8840817 … 1.3777062 1.3775196 1.3774453]
[1.8843104 1.88426 1.8841335 … 1.3776302 1.377445 1.3773713]
[1.884436 1.8843869 1.884263 … 1.3774408 1.3772594 1.3771874]…
Can someone suggest me how can i dequantise the uint16 tensor to match the ground truth floating point tensor.
Issue is don’t how how the fast-depth pre compiled model which i am using was quantised.