SegLD: Achieving universal, zero-shot and open-vocabulary segmentation through multimodal fusion via latent diffusion processes

Keyphrases

• Diffusion Process 100%
• Multimodal Fusion 100%
• Zero-shot 100%
• Open-vocabulary 100%
• Latent Diffusion 100%
• Segmentation Framework 85%
• Category Labels 42%
• Training Session 28%
• Image Category 28%
• Text Query 28%
• Image Caption 28%
• Unseen 28%
• Fully-supervised 28%
• Specific Task 14%
• Image Information 14%
• Training Set 14%
• Vocabulary Learning 14%
• Loss Function 14%
• Binary Mask 14%
• Label Information 14%
• Text Types 14%
• Segmentation Problem 14%
• Segmentation Performance 14%
• Text Input 14%
• Multi-scale Features 14%
• Transformer Encoder 14%
• Universal Applicability 14%
• Task Similarity 14%
• Text-to-image 14%
• Evaluation Metrics 14%
• COCO Dataset 14%
• Cityscapes Dataset 14%
• Segmentation Mask 14%
• Contrastive Loss 14%
• Classification Loss 14%
• Unknown Categories 14%
• Category-agnostic 14%
• Supervised Segmentation 14%
• Panopticon 14%

Computer Science

• Diffusion Process 100%
• Training Session 50%
• Segmentation Performance 25%
• Segmentation Task 25%
• Evaluation Metric 25%
• Identify Category 25%
• Contrastive Loss 25%
• Experimental Result 25%
• Image Information 25%