Special Session 7

Special Session 7: Retrieval-Augmented Generation and Fine-Grained Object Analysis

Description: Fine-grained object retrieval and analysis remain cornerstone challenges in computer vision, pattern recognition, and multimodal AI. These tasks are critical for applications such as fashion recommendation, species identification, cross-modal search, and industrial automation. While advances in deep learning have improved semantic representation, emerging paradigms like Retrieval-Augmented Generation (RAG) now offer transformative potential by integrating retrieval systems with generative models. This fusion enables context-aware, interpretable, and scalable solutions for fine-grained tasks, particularly when handling unlabeled data, unseen domains, or multimodal interactions.

However, challenges persist in balancing retrieval precision with generative diversity, optimizing cross-modal alignment, and ensuring efficiency in real-world systems. This special session seeks to unite cutting-edge research on fine-grained retrieval, matching, ranking, and RAG-driven methodologies, bridging gaps between traditional retrieval systems and generative AI.

We invite submissions that advance the theory, algorithms, and applications of fine-grained object analysis and retrieval-augmented systems. Emphasis will be placed on innovations that address multi-level semantics, cross-modal alignment, and domain adaptation, with a focus on scalability and real-world utility. Submissions exploring novel integrations of retrieval and generation (RAG) are particularly encouraged.

Session organizers
Assoc. Prof. Ying Li, Nanjing Normal University, China
Assoc. Prof. Tao Yao, Ludong University & Southwest Jiaotong University, China

The topics of interest include, but are not limited to:
• Fine-grained object retrieval and ranking in images, video, 3D data, or cross-modal contexts (text-to-image, image-to-text).
• Retrieval-Augmented Generation (RAG) for enhancing fine-grained analysis, including context-aware feature synthesis and hybrid retrieval-generation pipelines.
• Adversarial/generative models for robust representation learning in low-data or open-world scenarios.
• Self-supervised, meta-, and unsupervised learning for adapting retrieval/RAG systems to unseen domains or classes.
• Manifold learning, graph neural networks, and transformers for modeling complex semantic hierarchies in fine-grained data.
• Efficient indexing, real-time retrieval, and lightweight RAG architectures for large-scale applications.
• Cross-modal alignment and multimodal fusion techniques for joint retrieval and generation.
• Novel datasets, benchmarks, and evaluation metrics for fine-grained retrieval and RAG systems.
• Applications in e-commerce, healthcare, biodiversity monitoring, autonomous systems, and augmented reality.

Submission method
Submit your Full Paper (no less than 8 pages) or your paper abstract-without publication (200-400 words) via Online Submission System, then choose Special Session 7 (Retrieval-Augmented Generation and Fine-Grained Object Analysis)
Template Download

Introduction of session organizers