Researchers have developed SAM+D, a novel framework designed to adapt 2D foundation models like SAM and SAM2 for 3D volumetric and 4D spatiotemporal segmentation tasks. This parameter-efficient approach introduces two lightweight modules, Depth-Routed LoRA and Depth Shift Modules, which enable cross-slice feature exchange and spatially adaptive updates while keeping the majority of pre-trained parameters frozen. SAM+D achieves competitive results on medical imaging benchmarks for 3D segmentation and cell tracking datasets for 4D segmentation, utilizing significantly fewer trainable parameters than existing methods. AI
IMPACT Enables more efficient and effective 3D and 4D segmentation for medical imaging and biological research.
RANK_REASON The item is an academic paper detailing a new method for adapting existing models. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
