![图片](https://image.jiqizhixin.com/uploads/editor/a8ee9fe3-414f-456e-8da9-a5aded678f25/640.png)
论文地址:https://arxiv.org/pdf/2211.12254.pdf 论文主页:https://spinnerf3d.github.io/
![图片](https://image.jiqizhixin.com/uploads/editor/7c506498-f044-4cdf-a7b2-3422e8bc409e/640.png)
一个完整的 3D 场景操作过程,从用户交互的对象选择开始,到 3D 修复的 NeRF 场景结束; 将二维的分割模型扩展到多视图情况,能够从稀疏注释中恢复出具有三维一致的掩码; 确保视图一致性和感知合理性,一种新的基于优化的三维修复公式,利用二维图像修复; 一个新的用于三维编辑任务评估的数据集,包括相应的操作后的 Groud Truth。
![图片](https://image.jiqizhixin.com/uploads/editor/a48fa5ef-dc8c-4692-b293-09779b44874c/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/3950246d-04fd-412d-bf82-6a2c1260f6c0/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/357884ed-b8ec-499d-b776-18b6da4983af/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/83c57dea-be03-4deb-b275-abe680d63933/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/0067fce9-80f3-43a2-9ae9-9b53b9040d57/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/32a0edd2-1cb8-49fe-847c-d49301175ecf/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/0df72586-5f36-4ff9-9c4f-c56c2a5e50e4/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/cddbf934-24ee-4d8b-8768-abe742a18cdc/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/3897ea06-6b44-4d2e-a9bf-9cf63510d484/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/39f8f472-dfbb-4613-89a5-c4746407cf97/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/53ad90d8-bef8-4eb8-b17c-37641ae2f6f1/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/1374e36c-7afc-411f-831a-f254628bc791/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/6d856219-a674-4c5d-b4da-8a3960cca04a/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/b81120ac-d0cf-48f0-a569-df86602b5f8f/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/fc92947e-7a2b-4402-aa5a-3d8e0b7ccb3d/640.png)
![图片](https://image.jiqizhixin.com/uploads/editor/a280fbe6-d89f-4a76-892e-ecd62e100c7e/640.png)
Auto Byte
专注未来出行及智能汽车科技
微信扫一扫获取更多资讯
Science AI
关注人工智能与其他前沿技术、基础学科的交叉研究与融合发展
微信扫一扫获取更多资讯
神经辐射场(Neural Radiance Fields,简称 NeRF)已经成为一种流行的新视角合成方法。尽管 NeRF 迅速适应了更广泛的应用领域,但直观地编辑 NeRF 场景仍然是一个待解决的挑战。其中一个重要的编辑任务是从 3D 场景中移除不需要的对象,以使替换区域在视觉上是合理的,并与其上下文保持一致。本文提出了一种新颖的 3D 修复方法来解决这些挑战。