Perception across modalities, sensors, and scales.
Current programs include hierarchical transformers for geospatial tiles, temporal tracking under heavy occlusion, and self-supervised pre-training on multi-sensor industrial corpora. The lab publishes regularly and feeds weights and recipes into product channels.
Shared capture stages, calibration benches, drone and satellite data contracts, and the common Research Frameworks experiment OS. Researchers can move from notebook to multi-node training without rebuilding plumbing.
Manufacturing, smart cities, agriculture, and defence programs provide hard problems and deployment feedback. The lab measures success by field accuracy under distribution shift, not only clean benchmarks.
Vision Lab shares compute, evaluation, and publication tooling with every other Zansoc lab — so discoveries compound across the stack.
Visit all labs