arXiv Computer Vision research summaries for March 18, 2024.

Today's Research Themes (AI-Generated):

• Fed3DGS offers a scalable 3D reconstruction framework using federated learning to alleviate the load on centralized servers.

• Siamese Learning with SiamGTR provides a novel framework for Weakly-Supervised Video Paragraph Grounding without temporal annotations.

• Collage Prompting introduces a cost-effective image recognition strategy using GPT-4V that optimizes image arrangement for accuracy.

• MotionS is a novel generative pipeline for creating diverse, stylized motion across different character structures using cross-modality style prompts.

• SmartRefine presents an adaptive refinement strategy for efficient and scenario-specific motion prediction in dynamic, mixed human-robot environments.