SceneNet
收藏资源简介:
“标记合成室内场景的存储库。确保您在浏览器中启用了 WebGL 以查看 3D 模型。我们还在研究使用我们的模拟退火算法以 3D 场景的形式生成无限数据。我们还自动化了使用archivetextures和opensurfaces对这些场景进行纹理化。我们越来越多地看到这些场景的使用超越了标准的计算机视觉问题,例如语义分割、光流、3D场景重建等,到现在的物理场景理解和深度强化学习与代理与其3D交互环境。”
A repository for annotated synthetic indoor scenes. Please ensure WebGL is enabled in your browser to visualize the 3D models. We are also investigating the generation of infinite datasets in the form of 3D scenes using our simulated annealing algorithm. We have also automated the texturing of these scenes using archivetextures and opensurfaces. We have observed a growing range of applications for these scenes, extending beyond standard computer vision tasks such as semantic segmentation, optical flow, and 3D scene reconstruction, to now include physical scene understanding, deep reinforcement learning, and AI agent interaction with their 3D environments.

- SceneNet数据集首次发表,由NVIDIA研究团队提出,旨在为计算机视觉领域提供大规模的室内场景合成数据。
- SceneNet RGB-D版本发布,增加了深度信息,进一步提升了数据集的应用价值,特别是在三维重建和机器人导航领域。
- SceneNet数据集在多个国际会议和期刊上被广泛引用,成为室内场景理解和合成研究的重要基准。
- SceneNet数据集的应用扩展到自动驾驶和增强现实领域,展示了其在不同场景下的通用性和灵活性。
- SceneNet数据集的更新版本发布,增加了更多的场景类型和物体类别,进一步丰富了数据集的内容和多样性。
- 1SceneNet: Understanding Real World Indoor Scenes with Synthetic DataUniversity of California, Berkeley · 2016年
- 2SceneNet RGB-D: Can 5M Synthetic Images Beat Generic ImageNet Pre-training on Indoor Segmentation?University of California, Berkeley · 2018年
- 3Synthetic Data for Deep LearningUniversity of California, Berkeley · 2019年
- 4Learning to Segment Indoor Scenes from Synthetic DataUniversity of California, Berkeley · 2018年
- 5Synthetic Data for Training Deep Learning Models: A SurveyUniversity of California, Berkeley · 2020年



