Zero-BEV: Zero-shot Projection of Any First-Person Modality to BEV Maps
This enables zero-shot BEV mapping for various modalities in robotics, such as self-driving vehicles, addressing a bottleneck in existing methods that rely on depth or supervised training.
The paper tackles the problem of projecting first-person observations to bird's-eye view (BEV) maps without requiring depth information or modality-specific training, achieving this through a method that disentangles geometric projection from modality transformation and outperforms baselines like monocular depth estimation.
Bird's-eye view (BEV) maps are an important geometrically structured representation widely used in robotics, in particular self-driving vehicles and terrestrial robots. Existing algorithms either require depth information for the geometric projection, which is not always reliably available, or are trained end-to-end in a fully supervised way to map visual first-person observations to BEV representation, and are therefore restricted to the output modality they have been trained for. In contrast, we propose a new model capable of performing zero-shot projections of any modality available in a first person view to the corresponding BEV map. This is achieved by disentangling the geometric inverse perspective projection from the modality transformation, eg. RGB to occupancy. The method is general and we showcase experiments projecting to BEV three different modalities: semantic segmentation, motion vectors and object bounding boxes detected in first person. We experimentally show that the model outperforms competing methods, in particular the widely used baseline resorting to monocular depth estimation.