SegMAN blends State Space Models and Neighborhood Attention within a hybrid encoder-decoder for semantic segmentation, balancing long-range context with local detail. It achieves competitive mIoU on ADE20K with models from 6.4M to 92.6M parameters.
DAAAM builds real-time 3D dynamic scene graphs using foundation models like SAM and VLMs, targeting large-scale robot mapping with semantic and spatio-temporal memory.
