HumanEdit: A High-Quality Human-Rewarded Dataset for Instruction-based Image Editing
Jinbin Bai, Wei Chow, Ling Yang, Xiangtai Li, Juncheng Li, Hanwang Zhang, Shuicheng Yan
2024-12-06
Summary
This paper talks about HumanEdit, a new dataset created to improve instruction-based image editing by using high-quality images and detailed human feedback to guide the editing process.
What's the problem?
Many existing datasets for image editing do not include enough human feedback, which makes it difficult for AI models to understand what people really want when editing images. This lack of quality control can lead to poor results that don’t meet user expectations.
What's the solution?
The authors developed HumanEdit, which consists of 5,751 carefully curated images and detailed editing instructions. They employed human annotators to create data pairs and provide feedback, ensuring that the dataset reflects real-world editing needs. The dataset includes six types of editing instructions—Action, Add, Counting, Relation, Remove, and Replace—covering a wide range of scenarios. Additionally, all images come with masks to define the areas that need editing, and some instructions allow for editing without masks.
Why it matters?
This research is important because it sets a new standard for creating high-quality datasets for image editing. By focusing on human feedback and diverse editing tasks, HumanEdit can help improve the performance of AI models in real-world applications, making it easier for users to achieve the desired results when editing images.
Abstract
We present HumanEdit, a high-quality, human-rewarded dataset specifically designed for instruction-guided image editing, enabling precise and diverse image manipulations through open-form language instructions. Previous large-scale editing datasets often incorporate minimal human feedback, leading to challenges in aligning datasets with human preferences. HumanEdit bridges this gap by employing human annotators to construct data pairs and administrators to provide feedback. With meticulously curation, HumanEdit comprises 5,751 images and requires more than 2,500 hours of human effort across four stages, ensuring both accuracy and reliability for a wide range of image editing tasks. The dataset includes six distinct types of editing instructions: Action, Add, Counting, Relation, Remove, and Replace, encompassing a broad spectrum of real-world scenarios. All images in the dataset are accompanied by masks, and for a subset of the data, we ensure that the instructions are sufficiently detailed to support mask-free editing. Furthermore, HumanEdit offers comprehensive diversity and high-resolution 1024 times 1024 content sourced from various domains, setting a new versatile benchmark for instructional image editing datasets. With the aim of advancing future research and establishing evaluation benchmarks in the field of image editing, we release HumanEdit at https://huggingface.co/datasets/BryanW/HumanEdit.