- I saw several examples of VAE implementations.
- In VAE we use loss with 2 parts: MSE and KLDivLoss
- In all the examples I saw, they wrote the VAE loss (MSE+KL LOSS) in their own implementation and didn’t used:
torch.nn.KLDivLoss
One example can be found here:
https://github.com/AntixK/PyTorch-VAE/blob/master/models/vanilla_vae.py
Why did they implement the part of the KLDivLoss and didn’t use torch.nn.KLDivLoss
?