Torch.onnx support upsampling or convolutional transpose

I want to deploy semantic segmentation model. Currently onnx does not support upsampling or convolutional transpose. What are the alternatives for deploying the semantic segmentation?