single-view-3d

📚 NeurIPS 2024 - Single-view 3D Reconstruction Paper List

📄 Paper List

---

1. Hamba: Single-view 3D Hand Reconstruction with Graph-guided Bi-Scanning Mamba

Authors: Haoye Dong, Aviral Chharia, Wenbo Gou, Francisco Vicente Carrasco, Fernando De la Torre
Affiliations: Carnegie Mellon University

Highlights:

• 提出了一种结合图引导状态空间模型的 Mamba 架构，用于高效建模手部关节之间的空间关系。
• 设计了 Graph-guided State Space (GSS) 模块，使用比基于注意力的方法少 88.5% 的 token。
• 在 FreiHAND 数据集上实现了 PA-MPVPE 5.3mm 和 F@15mm 0.992 的性能，位列两个 3D 手部重建排行榜第一。:contentReference[oaicite:21]{index=21}

🔗 Resources

📄 Paper PDF | 🔗 OpenReview | 🎥 Presentation Video

🔍 Keywords: Single-view 3D Hand Reconstruction, Graph-guided Mamba, State Space Model
🗂️ Dataset: FreiHAND

---

2. Geometry Cloak: Preventing TGS-based 3D Reconstruction from Copyrighted Images

Authors: Qi Song, Ziyuan Luo, Ka Chun Cheung, Simon See, Renjie Wan
Affiliations: [机构信息待补充]

Highlights:

• :contentReference[oaicite:23]{index=23}
• :contentReference[oaicite:26]{index=26}:contentReference[oaicite:28]{index=28}

🔗 Resources

📄 Paper Abstract

🔍 Keywords: Single-view 3D Reconstruction, Triplane Gaussian Splatting, Image Protection
🗂️ Dataset: [数据集信息待补充]

---

3. Physically Compatible 3D Object Modeling from a Single Image

Authors: Minghao Guo, Bohan Wang, Pingchuan Ma, Tianyuan Zhang, Crystal Elaine Owens, Chuang Gan, Joshua B. Tenenbaum, Kaiming He, Wojciech Matusik
Affiliations: IBM Research, MIT, Meta AI

Highlights:

• 提出了一种计算框架，将单张图像转换为具有物理兼容性的 3D 物体模型。
• 通过显式分解物体的机械属性、外部力和静止形状几何，并通过静力平衡将其联系起来，确保优化后的物理形状表现出期望的物理行为。:contentReference[oaicite:33]{index=33}

🔗 Resources

📄 Paper Abstract

🔍 Keywords: Single-view 3D Reconstruction, Physical Compatibility, Static Equilibrium
🗂️ Dataset: Objaverse

---

4. L4GM: Large 4D Gaussian Reconstruction Model

Authors: Jiawei Ren, Cheng Xie, Ashkan Mirzaei, Hanxue Liang, Xiaohui Zeng, Karsten Kreis, Ziwei Liu, Antonio Torralba, Sanja Fidler, Seung Wook Kim, Huan Ling
Affiliations: [机构信息待补充]

Highlights:

• 提出了 L4GM，第一个从单视图视频输入生成动画对象的 4D 大型重建模型。
• 该模型在一次前馈传递中生成每帧的 3D 高斯点云表示，并通过插值模型将其上采样到更高的帧率，实现时间上的平滑性。:contentReference[oaicite:38]{index=38}

🔗 Resources

📄 Paper PDF

🔍 Keywords: Single-view 3D Reconstruction, 4D Gaussian Splatting, Dynamic Objects
🗂️ Dataset: Objaverse

---

请注意：由于目前 NeurIPS 2025 的完整论文列表尚未公开，以上为 NeurIPS 2024 中的代表性论文。建议您定期访问 NeurIPS 2025 官方网站 以获取最新的论文信息和资源。

This site is open source. Improve this page.