October 12, 2025Open Access

Scaling Transformer-Based Novel View Synthesis Models with Token Disentanglement and Synthetic Data

View Full Paper

Ask AI

Bookmark

Our method significantly improves reconstruction quality using token disentanglement within transformer models.
By incorporating synthetic training data from diffusion models, we achieve better generalization across unseen domains.
Transformers enhanced with token disentanglement outperform existing models on various benchmarks, reducing computational costs.
We identify and address artifacts in synthetic data generation as a key challenge impacting reconstruction performance.

Discussion

No takes yet. Share an insight, caveat, or question.

Innovative approach demonstrates improved reconstruction quality in novel view synthesis, highlighting token disentanglement benefits.

Nair et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark