SparseGNV: Generating Novel Views of Indoor Scenes with Sparse Input Views

TL;DR

SparseGNV is a novel framework that generates photorealistic indoor scene views from sparse inputs by combining 3D geometry, transformer-based decoding, and learned priors, outperforming existing methods.

Contribution

It introduces a three-module learning framework integrating neural point clouds, transformers, and image reconstruction for efficient view synthesis.

Findings

Outperforms state-of-the-art methods on real-world indoor scenes.

Efficient feed-forward generation of novel views of unseen scenes.

Effective use of 3D structures and generative models for view synthesis.

Abstract

We study to generate novel views of indoor scenes given sparse input views. The challenge is to achieve both photorealism and view consistency. We present SparseGNV: a learning framework that incorporates 3D structures and image generative models to generate novel views with three modules. The first module builds a neural point cloud as underlying geometry, providing contextual information and guidance for the target novel view. The second module utilizes a transformer-based network to map the scene context and the guidance into a shared latent space and autoregressively decodes the target view in the form of discrete image tokens. The third module reconstructs the tokens into the image of the target view. SparseGNV is trained across a large indoor scene dataset to learn generalizable priors. Once trained, it can efficiently generate novel views of an unseen indoor scene in a feed-forward manner. We evaluate SparseGNV on both real-world and synthetic indoor scenes and demonstrate that it outperforms state-of-the-art methods based on either neural radiance fields or conditional image generation.