Jetson Nano Face Recognition


I’m trying to run a face recognition model on Jetson Nano with Open-CV Fast Math Enabled the range of the metric that is calculated using face distance is 0 to 1 but with Fast Math On the values are off by large number which is returning -835.66 e^34
With Open-CV Fast Math disabled it’s getting killed


Just want to confirm first.

Do you use OpenCV or the face recognition python module, which is simply based on the dlib library?
There is a known issue in dlib that generate NAN or very small number, similar to your description.

The issue is caused by cuDNN regression and here is a workaround for it: