How does DC-AE 1.5 compare with VA-VAE for handling large visual feature dimensions?

Hi, thank you for the amazing work on DC-AE — it’s really impressive!

I’m curious about how DC-AE 1.5 compares with [VA-VAE](https://arxiv.org/html/2501.01423v3), especially since VA-VAE also addresses the optimization difficulty that arises in latent diffusion models when the visual tokenizer’s feature dimension increases.

Thanks again for sharing your great work!