Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2504.02817 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908476349874176 |
|---|---|
| author | Deng, Kangle Liu, Hsueh-Ti Derek Zhu, Yiheng Sun, Xiaoxia Shang, Chong Bhat, Kiran Ramanan, Deva Zhu, Jun-Yan Agrawala, Maneesh Zhou, Tinghui |
| author_facet | Deng, Kangle Liu, Hsueh-Ti Derek Zhu, Yiheng Sun, Xiaoxia Shang, Chong Bhat, Kiran Ramanan, Deva Zhu, Jun-Yan Agrawala, Maneesh Zhou, Tinghui |
| contents | Many 3D generative models rely on variational autoencoders (VAEs) to learn compact shape representations. However, existing methods encode all shapes into a fixed-size token, disregarding the inherent variations in scale and complexity across 3D data. This leads to inefficient latent representations that can compromise downstream generation. We address this challenge by introducing Octree-based Adaptive Tokenization, a novel framework that adjusts the dimension of latent representations according to shape complexity. Our approach constructs an adaptive octree structure guided by a quadric-error-based subdivision criterion and allocates a shape latent vector to each octree cell using a query-based transformer. Building upon this tokenization, we develop an octree-based autoregressive generative model that effectively leverages these variable-sized representations in shape generation. Extensive experiments demonstrate that our approach reduces token counts by 50% compared to fixed-size methods while maintaining comparable visual quality. When using a similar token length, our method produces significantly higher-quality shapes. When incorporated with our downstream generative model, our method creates more detailed and diverse 3D content than existing approaches. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_02817 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Efficient Autoregressive Shape Generation via Octree-Based Adaptive Tokenization Deng, Kangle Liu, Hsueh-Ti Derek Zhu, Yiheng Sun, Xiaoxia Shang, Chong Bhat, Kiran Ramanan, Deva Zhu, Jun-Yan Agrawala, Maneesh Zhou, Tinghui Computer Vision and Pattern Recognition Many 3D generative models rely on variational autoencoders (VAEs) to learn compact shape representations. However, existing methods encode all shapes into a fixed-size token, disregarding the inherent variations in scale and complexity across 3D data. This leads to inefficient latent representations that can compromise downstream generation. We address this challenge by introducing Octree-based Adaptive Tokenization, a novel framework that adjusts the dimension of latent representations according to shape complexity. Our approach constructs an adaptive octree structure guided by a quadric-error-based subdivision criterion and allocates a shape latent vector to each octree cell using a query-based transformer. Building upon this tokenization, we develop an octree-based autoregressive generative model that effectively leverages these variable-sized representations in shape generation. Extensive experiments demonstrate that our approach reduces token counts by 50% compared to fixed-size methods while maintaining comparable visual quality. When using a similar token length, our method produces significantly higher-quality shapes. When incorporated with our downstream generative model, our method creates more detailed and diverse 3D content than existing approaches. |
| title | Efficient Autoregressive Shape Generation via Octree-Based Adaptive Tokenization |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2504.02817 |