Hi @dusty-nv ,

Thanks for all the work you do on the inference repo. I realized you are using this repo to provide the models for the segmentation.

I was wondering if there was a quick and simple tutorial somewhere that explains the best way to use this repo in order to train on custom datasets and then use with the jetsoninfernce libs.

Thanks!

Hi There,

I'm having some issues trying to use this to train a custom image segmentation model and having some issues. I see the custom fork for torchvision v.0.3.0 but just wondering which version of torch this was written for.

Thanks

I am having issue in converting a custom trained segmentation network using pytorch on resnet101 network to convert to ONNX format using the onnx_export.py utility. It does not accept the export_onnx=True in the models.segmentation.__dict__[arch](num_classes=num_classes, aux_loss=None, pretrained=False, export_onnx=True). Is it safe OK to remove this parameter? Also how to load the custom trained onnx file for model inference? @dusty-nv

Hi :)
Why is the network limited to 21 classes?

Hello,
Thank you for the work you provided. I am now trying to use the network fcn-resnet18-deepscene-576x320 in my research. I run into trouble because the environment I am testing in has more shadows. So I need to fine tune the model with a dataset that I gathered myself.
It may sound silly, but I wonder how can I retrain your model? (or better said how to fine tune the model with extra dataset?) (I use Python 3.6 mainly)
I appreciate your help.

Best Regards,
erKarim

Or can we use the latest torchvision models instead?

Hi @dusty-nv ,

I want to load and test pre-trained segmentation models such as FCN-ResNet18-MHP, FCN-ResNet18-DeepScene on my system with python script e.g. test.py before using on Jetson.

How can we achieve the same?

Please let me know if any further information w.r.t is needed.

Thanks!

Hi,
I have downloaded VOCtrainval_11-May-2012 dataset.
Is there any pre-process for above data to run code available in this repo.
What is the folder format should be provided.
Available code is only for training or can do inference also.
I am a C++ guy. Not aware of python coding.
Kindly do help.

Thank You.