ActiveSAM: Image-Conditional Class Pruning for Fast and Accurate Open-Vocabulary Segmentation
Researchers have developed ActiveSAM, a novel framework designed to enhance the efficiency and accuracy of open-vocabulary semantic segmentation using the Segment Anything Model 3 (SAM 3). This training-free, zero-shot approach optimizes the segmentation process by identifying an active subset of classes relevant to each image, thereby reducing computational load. ActiveSAM demonstrates significant improvements in speed and accuracy, outperforming existing state-of-the-art methods like SegEarth-OV3 and showing strong robustness against image corruption, making it suitable for real-world applications. AI
IMPACT ActiveSAM's efficiency gains could accelerate deployment of open-vocabulary segmentation in real-world applications like autonomous driving.