Bilinear sampler at loss calculation

vision
1

I used differentiable bilinear sampler from STN during loss calculation in my model as below:

mask = model(input)
output = bilinear_sampler(input, mask)
loss = loss_cal(input, output)

Will the gradient flow automatically when loss.backward() is called?

model.optimizer.zerograd()
loss.backward()
model.optimizer.steps()

Should I write backward() function separately for this model? Thank you.