Export PyTorch model to ONNX model

[WillBao33]

I trained a Retinanet model using the existing config file from Detectron2 with the following command:
python train.py --output_path ./retina --yaml COCO-Detection/retinanet_R_101_FPN_3x.yaml --resume --max_iter 20000 --lr_rate 1e-5 --batch_size 16 --iou 0.4
I wonder how to use the caffe2_export.py to export the PyTorch file (.pth) to the ONNX file. What's the command that I should use and do I need to parse any arguments in the command?
If this is not the right way to export the ONNX, could you suggest some methods to do it?
Thanks

[thiagocrepaldi]

I am working on adding support to ONNX export (with and without caffe2) #4120 and #4153

Hopefully it gets accepted soon