SPSRL: Open-vocabulary semantic segmentation with spatial prior and semantic relation learning

Puntos clave

• The method achieves improved semantic segmentation accuracy through open-vocabulary approaches, integrating spatial prior and semantic relations.
• Accuracy increases significantly for diverse image datasets, resulting in a notable reduction in classification errors.
• Analysis using deep learning algorithms uncovers strong associations between spatial features and semantic categories, enhancing performance.
• The findings highlight the potential for open-vocabulary techniques to enable flexible image interpretation in various fields.