InstructDiffusion: A Generalist Modeling Interface for Vision Tasks

TL;DR

InstructDiffusion introduces a versatile diffusion-based framework that aligns diverse vision tasks with human instructions, enabling flexible, interactive image manipulation and outperforming prior methods on various datasets.

Contribution

It presents a novel, unified diffusion-based approach that models vision tasks as human-guided image editing, handling both understanding and generative tasks with improved generalization.

Findings

Handles a variety of vision tasks including segmentation and editing

Outperforms prior methods on novel datasets

Capable of managing unseen tasks

Abstract

We present InstructDiffusion, a unifying and generic framework for aligning computer vision tasks with human instructions. Unlike existing approaches that integrate prior knowledge and pre-define the output space (e.g., categories and coordinates) for each vision task, we cast diverse vision tasks into a human-intuitive image-manipulating process whose output space is a flexible and interactive pixel space. Concretely, the model is built upon the diffusion process and is trained to predict pixels according to user instructions, such as encircling the man's left shoulder in red or applying a blue mask to the left car. InstructDiffusion could handle a variety of vision tasks, including understanding tasks (such as segmentation and keypoint detection) and generative tasks (such as editing and enhancement). It even exhibits the ability to handle unseen tasks and outperforms prior methods on novel datasets. This represents a significant step towards a generalist modeling interface for vision tasks, advancing artificial general intelligence in the field of computer vision.