CVDec 22, 2024

DreamOmni: Unified Image Generation and Editing

arXiv:2412.17098v223 citationsh-index: 15CVPR
Originality Incremental advance
AI Analysis

This addresses the problem of fragmented model deployment for image generation and editing in computer vision, offering a more usable and synergistic approach, though it is incremental as it builds on existing T2I frameworks.

The paper tackles the lack of unified multitasking in computer vision by introducing DreamOmni, a model that integrates text-to-image generation with various editing tasks, resulting in significantly boosted editing performance through joint training and a synthetic data pipeline.

Currently, the success of large language models (LLMs) illustrates that a unified multitasking approach can significantly enhance model usability, streamline deployment, and foster synergistic benefits across different tasks. However, in computer vision, while text-to-image (T2I) models have significantly improved generation quality through scaling up, their framework design did not initially consider how to unify with downstream tasks, such as various types of editing. To address this, we introduce DreamOmni, a unified model for image generation and editing. We begin by analyzing existing frameworks and the requirements of downstream tasks, proposing a unified framework that integrates both T2I models and various editing tasks. Furthermore, another key challenge is the efficient creation of high-quality editing data, particularly for instruction-based and drag-based editing. To this end, we develop a synthetic data pipeline using sticker-like elements to synthesize accurate, high-quality datasets efficiently, which enables editing data scaling up for unified model training. For training, DreamOmni jointly trains T2I generation and downstream tasks. T2I training enhances the model's understanding of specific concepts and improves generation quality, while editing training helps the model grasp the nuances of the editing task. This collaboration significantly boosts editing performance. Extensive experiments confirm the effectiveness of DreamOmni. The code and model will be released.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes