VoxFormer: Sparse Voxel Transformer for Camera-based 3D Semantic Scene Completion
This work addresses the challenge of enabling AI systems to infer complete 3D scenes from limited 2D visual data, which is crucial for applications like autonomous driving and robotics, though it is an incremental advancement building on existing methods.
The paper tackles the problem of 3D semantic scene completion from 2D images by proposing VoxFormer, a Transformer-based framework that uses a two-stage design with sparse voxel queries and densification, achieving state-of-the-art results with relative improvements of 20.0% in geometry and 18.1% in semantics on SemanticKITTI.
Humans can easily imagine the complete 3D geometry of occluded objects and scenes. This appealing ability is vital for recognition and understanding. To enable such capability in AI systems, we propose VoxFormer, a Transformer-based semantic scene completion framework that can output complete 3D volumetric semantics from only 2D images. Our framework adopts a two-stage design where we start from a sparse set of visible and occupied voxel queries from depth estimation, followed by a densification stage that generates dense 3D voxels from the sparse ones. A key idea of this design is that the visual features on 2D images correspond only to the visible scene structures rather than the occluded or empty spaces. Therefore, starting with the featurization and prediction of the visible structures is more reliable. Once we obtain the set of sparse queries, we apply a masked autoencoder design to propagate the information to all the voxels by self-attention. Experiments on SemanticKITTI show that VoxFormer outperforms the state of the art with a relative improvement of 20.0% in geometry and 18.1% in semantics and reduces GPU memory during training to less than 16GB. Our code is available on https://github.com/NVlabs/VoxFormer.