TOSS:High-quality Text-guided Novel View Synthesis from a Single Image

TL;DR

TOSS leverages text guidance and fine-tuned diffusion models to produce high-quality, controllable, and multiview-consistent novel view synthesis from a single image, addressing limitations of previous image-to-image translation methods.

Contribution

The paper introduces TOSS, a novel framework that integrates text semantics with diffusion models for improved single-image NVS with explicit control and detail preservation.

Findings

TOSS outperforms Zero-1-to-3 in plausibility and multiview consistency.

Text guidance enhances control over generated views.

Architecture improvements improve pose accuracy and detail retention.

Abstract

In this paper, we present TOSS, which introduces text to the task of novel view synthesis (NVS) from just a single RGB image. While Zero-1-to-3 has demonstrated impressive zero-shot open-set NVS capability, it treats NVS as a pure image-to-image translation problem. This approach suffers from the challengingly under-constrained nature of single-view NVS: the process lacks means of explicit user control and often results in implausible NVS generations. To address this limitation, TOSS uses text as high-level semantic information to constrain the NVS solution space. TOSS fine-tunes text-to-image Stable Diffusion pre-trained on large-scale text-image pairs and introduces modules specifically tailored to image and camera pose conditioning, as well as dedicated training for pose correctness and preservation of fine details. Comprehensive experiments are conducted with results showing that our proposed TOSS outperforms Zero-1-to-3 with more plausible, controllable and multiview-consistent NVS results. We further support these results with comprehensive ablations that underscore the effectiveness and potential of the introduced semantic guidance and architecture design.