Click2Graph: Interactive Panoptic Video Scene Graphs from a Single Click

The Segment Anything Model 2 (SAM2) has shown impressive universal segmentation capabilities on natural images, but its performance on ultrasound images is hindered by domain disparities. To tackle this issue, UniUltra is proposed, featuring a context-edge hybrid adapter (CH-Adapter) for enhanced ultrasound imaging perception and a deep-supervised knowledge distillation (DSKD) technique to facilitate effective deployment in clinical settings.

The proposed task of Collision Sound Source Segmentation (CS3) aims to identify and segment objects responsible for collision sounds in egocentric videos. This task addresses challenges such as cluttered visual scenes and brief interactions, utilizing a weakly-supervised method that leverages audio cues and foundation models like CLIP and SAM2. The focus on egocentric video allows for clearer sound identification despite visual complexity.

VideoSeg-R1 is a novel framework that integrates reinforcement learning into video object segmentation, overcoming limitations of traditional supervised methods. It features a decoupled architecture that combines referring image segmentation with video mask propagation, utilizing a hierarchical text-guided frame sampler, a reasoning model, and a segmentation-propagation stage. This approach enhances efficiency and accuracy in complex video reasoning tasks, achieving state-of-the-art performance across multiple benchmarks.

The Segment Anything Model 2 (SAM2) has been enhanced with the introduction of SAM2S, a model designed for surgical video segmentation. This development addresses challenges in long-term tracking and domain gaps in surgical scenarios by utilizing the SA-SV benchmark, which includes extensive spatio-temporal annotations. The model incorporates a diverse memory mechanism and temporal semantic learning to improve instrument and tissue tracking in surgical videos.