I am trying to run some face recognition encoding process on Jetson Nano. It seems to work fine (face_recognition.face_encoding) as long as num_jitters = 1 (by default). this is responsilbe to make samples of the image and then giving the encoding average values. However, when I set num_jitters to bigger values than 1 e.g. 100, the returned values become NaN. I though it might the package issue, but I ran the same coding on a laptop using CPU (no GPU), the results were not NaN for the same sampling (images). In addition, I noticed that when num_jitters is a small value like 2, then a few images might have encodings, however, most of them were not.
What does that mean? Why Jetson Nano (GPU) was not able to find the encodings whereas a CPU was?