Hello! I need help with using one model to calculate the loss for another model! Essentially I use model1 to generate a face image based on a real face image. model2 which is the media pipe face mesh calculates the face mesh for both the real and fake faces and does MSEloss on it. I want to then backpropagate that loss to model1. Because I run the MSE on the output of the face mesh, it won’t backpropagate the gradients of model1. How can I do this?
I don’t fully understand this claim so could you explain the issue in more detail, please?
MSELoss won’t detach the computation graph and is differentiable, so I don’t know what the root cause of the failure is.