Does TensorRT support pytorch Maskrcnn

Hi,

I am willing to convert a Pytorch Object Detection model from torchvision.models.detection, concretly the Maskrcnn or Fasterrcnn to a TensorRT model.
I know I can do it via torch2trt or via onnx model, but I am not sure if the models will be compatible.
For instance, the Pytorch maskrcnn model has a FrozenBatchNormalization2d() layer, but in the documentation I see just support for BatchNormalization()
Is it possible the conversion?

Documentation from here:

Thank you for your time

Hi,

Yes, but you have to create a custom plugin for any unsupported ops in the model.
Please refer to below sample:
https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/sampleUffMaskRCNN

Supported ONNX ops:
https://github.com/onnx/onnx-tensorrt/blob/master/operators.md

Thanks

1 Like