SegNet SUN

I download SUN(indoor scene) prototxt and caffemodel from https://github.com/alexgkendall/SegNet-Tutorial/blob/master/Example_Models/segnet_model_zoo.md and copy into Jetson Nano. It doesn’t work. How do I convert/modify for segnet-console?

PS. newbie in deep learning

Hi davidwu0709, the segnet-console program is compatible with segmentation models using FCN-Alexnet models (or derivatives like FCN-8S). Not sure that it will be able to use this other segmentation model you linked to due to custom layers that it appears to use (it says it uses a modified version of caffe). What you could do, is just run that GitHub code and not use it with segnet-console. It won’t be using TensorRT but you can at least run it.

Alternatively, what I would recommend is trying to follow the segmentation training tutorial with FCN-Alexnet, but using the dataset that you desire.

Hi,

According to this page(https://devblogs.nvidia.com/jetson-nano-ai-computing/), seems that it is possible to do training/re-training by using PyTorch on Jetson Nano, not using DIGITS. Could you kindly provide any information or tutorial about this? thanks in advance.

Hi, yes, I ran those using this example training code from PyTorch: https://github.com/pytorch/examples/tree/master/imagenet

You may need to mount a 2GB swapfile for it: https://support.rackspace.com/how-to/create-a-linux-swap-file/