Clarification on IM-GAN vs IM-AE about im-net HOT 2 CLOSED

We used a latent GAN in our paper, because it can be trained much faster than a common GAN.

In a common GAN framework:
random code -> Generator -> output shape.

In a latent GAN framework:
random code -> Generator -> output code -> Shape Decoder -> output shape.

Therefore you saw what you saw in the code.

This repo is dedicated for shape autoencoding and single view reconstruction on the 13 ShapeNet categories. Although you could train a GAN with the provided code, we do not provide pretrained weights.

If you need the weights, please go to the original implementation:
https://github.com/czq142857/implicit-decoder.

from im-net.

Thank you for the clarification, that makes sense.

from im-net.