Semi-transductive Learning for Generalized Zero-Shot Sketch-Based Image Retrieval

Sketch-based image retrieval (SBIR) is an attractive research area where freehand sketches are used as queries to retrieve relevant images. Existing solutions have advanced the task challenging zero-shot setting (ZS-SBIR), trained models tested on new classes without seen data. However, they prone overfitting under a realistic scenario when test data includes both and unseen classes. In this paper, we study generalized ZS-SBIR (GZS-SBIR) propose novel semi-transductive learning paradigm. Transductive performed modality explore potential distribution within classes, sketch sharing learned knowledge through semi-heterogeneous architecture. A hybrid metric strategy proposed establish semantics-aware ranking property calibrate joint embedding space. Extensive experiments conducted two large-scale benchmarks four evaluation metrics. The results show that our method superior over state-of-the-art competitors in GZS-SBIR task.