The paper introduces EgoISM-HOI, a new multimodal dataset for detecting Egocentric Human-Object Interactions in industrial scenarios. By leveraging synthetic data and multimodal signals, the proposed method significantly improves performance when tested on real-world data. The study highlights the benefits of using synthetic data to pre-train detection methods and compares them with state-of-the-art approaches.
The research addresses the lack of public datasets by creating a pipeline to generate synthetic images paired with annotations for hands and objects. It demonstrates how wearable devices can be used to monitor human-object interactions in industrial contexts. The study emphasizes the importance of understanding human-object interactions from a first-person perspective using synthetic generated data.
Furthermore, the paper discusses related works on datasets and methods for detecting human-object interactions from different perspectives. It also explores simulators and synthetic datasets that aid in generating labeled data automatically. The proposed approach stands out by focusing on realistic 3D reconstructions of environments and objects to create accurate synthetic interactions.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Rosario Leon... at arxiv.org 03-12-2024
https://arxiv.org/pdf/2306.12152.pdfDeeper Inquiries