**4D-DRESS: A 4D Dataset of Real-World Human Clothing With Semantic Annotations** **Authors:** Wenbo Wang **Affiliations:** Department of Computer Science, ETH Zürich; Max Planck Institute for Intelligent Systems, Tübingen **URL:** https://ait.ethz.ch/4d-dress **Overview:** 4D-DRESS is the first real-world 4D dataset of human clothing, capturing 64 human outfits in over 520 motion sequences. Each sequence includes high-quality 4D textured scans, vertex-level semantic labels, garment meshes, and fitted SMPL(-X) body meshes. The dataset aims to advance research in realistic clothing dynamics and simulation. **Key Features:** - **High-Quality 4D Textured Scans:** Captured using advanced multi-view volumetric capture systems. - **Vertex-Level Semantic Labels:** Annotate various clothing types (upper, lower, outer) with high accuracy. - **Garment Meshes and SMPL-X Body Meshes:** Provide detailed 3D representations for realistic simulations and reconstructions. **Challenges Addressed:** - **Real-World Clothing Dynamics:** Traditional synthetic datasets often lack realistic clothing dynamics and can fail to capture complex movements. - **Annotation and Segmentation:** annotating and segmenting extensive and complex 4D human scans is challenging. **Methods:** - **Semi-Automatic 4D Human Parsing Pipeline:** Combines human-in-the-loop processes with automation to accurately label 4D scans. - **Graph Cut Optimization:** Efficiently assigns vertex labels using a graph cut algorithm, incorporating multi-view and temporal consistency. **Evaluation:** - **Benchmarks for Clothing Simulation and Reconstruction:** Establishes evaluation benchmarks to assess the effectiveness of clothing simulation and reconstruction methods. - **Qualitative and Quantitative Analysis:** Demonstrates the dataset's utility through various experiments and ablation studies. **Contributions:** - **First Real-World 4D Human Clothing Dataset:** Provides a comprehensive resource for researchers. - **Semi-Automatic 4D Human Parsing Pipeline:** Offers an efficient method for data annotation. - **Evaluation Benchmarks:** Highlights the dataset's value for advancing research in clothing simulation and reconstruction. **Conclusion:** 4D-DRESS bridges the gap between synthetic and real-world datasets, providing high-quality 4D data for realistic clothing research. It is a valuable resource for advancing the field of lifelike human clothing in computer vision and graphics.**4D-DRESS: A 4D Dataset of Real-World Human Clothing With Semantic Annotations** **Authors:** Wenbo Wang **Affiliations:** Department of Computer Science, ETH Zürich; Max Planck Institute for Intelligent Systems, Tübingen **URL:** https://ait.ethz.ch/4d-dress **Overview:** 4D-DRESS is the first real-world 4D dataset of human clothing, capturing 64 human outfits in over 520 motion sequences. Each sequence includes high-quality 4D textured scans, vertex-level semantic labels, garment meshes, and fitted SMPL(-X) body meshes. The dataset aims to advance research in realistic clothing dynamics and simulation. **Key Features:** - **High-Quality 4D Textured Scans:** Captured using advanced multi-view volumetric capture systems. - **Vertex-Level Semantic Labels:** Annotate various clothing types (upper, lower, outer) with high accuracy. - **Garment Meshes and SMPL-X Body Meshes:** Provide detailed 3D representations for realistic simulations and reconstructions. **Challenges Addressed:** - **Real-World Clothing Dynamics:** Traditional synthetic datasets often lack realistic clothing dynamics and can fail to capture complex movements. - **Annotation and Segmentation:** annotating and segmenting extensive and complex 4D human scans is challenging. **Methods:** - **Semi-Automatic 4D Human Parsing Pipeline:** Combines human-in-the-loop processes with automation to accurately label 4D scans. - **Graph Cut Optimization:** Efficiently assigns vertex labels using a graph cut algorithm, incorporating multi-view and temporal consistency. **Evaluation:** - **Benchmarks for Clothing Simulation and Reconstruction:** Establishes evaluation benchmarks to assess the effectiveness of clothing simulation and reconstruction methods. - **Qualitative and Quantitative Analysis:** Demonstrates the dataset's utility through various experiments and ablation studies. **Contributions:** - **First Real-World 4D Human Clothing Dataset:** Provides a comprehensive resource for researchers. - **Semi-Automatic 4D Human Parsing Pipeline:** Offers an efficient method for data annotation. - **Evaluation Benchmarks:** Highlights the dataset's value for advancing research in clothing simulation and reconstruction. **Conclusion:** 4D-DRESS bridges the gap between synthetic and real-world datasets, providing high-quality 4D data for realistic clothing research. It is a valuable resource for advancing the field of lifelike human clothing in computer vision and graphics.