Encounter "aten::resize is not implemented" runtime error during backward of loss

I’m using C++ frontend API to implement the training loop.

  • “pred” are the the predictions of a 2D Unet model from a batch of training data.
  • “target” are the label of the training data.
  • “loss_f” is the torch::nn::BCEWithLogitsLoss().
  • “aten::resize is not implemented” runtime error appear at line 134.

Any idea to solve this issue?