Jetson inference image segmentation result different from DIGIT

I follow the sky-terrain segmentation example here(https://github.com/dusty-nv/jetson-inference), train a segnet with DIGIT in my PC and deploy the net with jetson inference. But i get different result between DIGIT and jetson-inference with the same input image. The segmented boundary is zigzag in jetson-inference while it is smooth in DIGIT.

  1. How can i get the same segmentation result in TX2?
  2. I notice that i delete the deconv and crop layers and on line 24 of deploy.prototxt, change pad: 100 to pad: 0. I guess that might be the reason for the difference. What’s that for?
  3. Are there any python script for inferring segnet?

@AastaLLL Could you please answer these? Thank you

Hi,

Please check this topic for more information:
https://devtalk.nvidia.com/default/topic/1021784/jetson-tx2/unable-to-get-segmentation-to-work-with-jetson-tx2/post/5201703/#5201703

Thanks.