ICCV 2022 Open Access Repository

From Fake News
Jump to: navigation, search

Zekun Hao, Arun Mallya, Serge Belongie, Ming-Yu Liu Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021 (pp. 14072-14082



Abstract



We present GANcraft, an unsupervised neural rendering framework for generating realistic images of large 3D block worlds like the ones created by Minecraft. Our approach employs an underlying semantic block world as input. Each block is given an associated semantic label like dirt, grass, and water. The world is represented as an ongoing volumetric function. We train our model to render consistent, view-consistent pictures for a camera controlled by a user. We designed a training technique that relies on adversarial and pseudo-ground truth training, in the absence of actual images from the block world. This stands in contrast to prior research on neural rendering for view synthesis, which requires ground truth images to determine the geometry of the scene and also to determine the appearance that is dependent on view. In addition to tracking the camera, GANcraft allows user control over both scene semantics and output style. Results from experiments compared to solid baselines prove the effectiveness of GANcraft in this new task of photorealistic 3D block synthesizing.

Mega Blog