Sign in to view source links and access this dataset
Description
HOI-Edit-44K is a large-scale dataset of 44,117 high-quality, paired examples for Human-Object Interaction (HOI) editing, created to address the scarcity of supervised data for this task. It was authored by jiuntian and released on Hugging Face in April 2026, with associated research presented at CVPR 2026. The dataset is designed to provide the necessary supervision for training models like OneHOI to perform identity-preserving HOI modifications.
Use Cases
Training generative models for identity-preserving human-object interaction editing based on the paired examples.
Benchmarking model performance on human-object interaction modification tasks based on the large-scale, supervised data.
Developing and evaluating methods for robust HOI scene manipulation based on the high-quality, paired examples.