BlendedMVS
收藏资源简介:
BlendedMVS是一个大规模的多视角立体网络训练数据集,由香港科技大学创建。该数据集包含超过17,000张高分辨率图像,覆盖城市、建筑、雕塑和小型物体等多种场景。数据集的创建过程涉及从精选场景的图像中恢复高质量的纹理网格,然后渲染这些网格模型以生成彩色图像和深度图。为了在训练中引入环境光照信息,渲染的彩色图像进一步与输入图像混合以生成训练输入。BlendedMVS数据集旨在解决多视角立体视觉任务中训练数据不足的问题,显著提高模型的泛化能力。
BlendedMVS is a large-scale multi-view stereo training dataset developed by The Hong Kong University of Science and Technology. It contains over 17,000 high-resolution images covering diverse scenarios including urban scenes, buildings, sculptures and small objects. The dataset creation workflow involves recovering high-quality textured meshes from images of carefully selected scenes, then rendering these mesh models to generate color images and depth maps. To incorporate environmental lighting information during training, the rendered color images are further blended with input images to produce training inputs. BlendedMVS aims to address the shortage of training data in multi-view stereo tasks and significantly enhances the generalization ability of models.




