ONNX to TorchScript conversion for quantized networks

There is a quantized network in ONNX format that I`d like to convert to TorchScript.
I know that such conversions are usually done backwards, from TS to ONNX, but I want to test my TS integration on some quantized models and I was told the easiest way is to use ONNX.

  1. Does ONNX use the same execution pipeline as TorchScript which I can interface via RegisterPass and RegisterOperators?
  2. If not, then how do I perform a ONNX → TorchScript conversion?
  3. If this is impossible, how do I acquire and trace a quantized, preferably asymmetric, ResNet50 model?