PCA+MM-VAE based multimodal data compression retrieval method

The rapid growth of multimodal data applications has introduced significant challenges in efficient storage and cross-modal retrieval. While existing approaches such as common subspace learning, deep feature extraction, and hashing techniques have made progress, they still face critical limitations in computational efficiency and retrieval performance. Recent advances using Vision-Language Pretrained models (e.g., CLIP with Dual Query Unification for Compositional Image Retrieval, DQU-CIR) have shown promise but remain computationally intensive.