Sherpa3D
Sherpa3D copied to clipboard
Is the structure loss and the semantic loss didn't actually work in your code?
The weight of your proposed loss(structure loss and semantic loss ) in your code changes like this?
And it actually works at step 1000.
It means that the structure loss and the semantic loss didn't actually work.
The author used simulated annealing. As the number of iterations increases, the impact becomes smaller.