Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2501.10834 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917896845787136 |
|---|---|
| author | Bonomo, Mirco Bianco, Simone |
| author_facet | Bonomo, Mirco Bianco, Simone |
| contents | Multimodal Large Language Models (MLLMs) have achieved notable performance in computer vision tasks that require reasoning across visual and textual modalities, yet their capabilities are limited to their pre-trained data, requiring extensive fine-tuning for updates. Recent researches have explored the use of In-Context Learning (ICL) to overcome these challenges by providing a set of demonstrating examples as context to augment MLLMs performance in several tasks, showing that many-shot ICL leads to substantial improvements compared to few-shot ICL. However, the reliance on numerous demonstrating examples and the limited MLLMs context windows presents significant obstacles. This paper aims to address these challenges by introducing a novel approach, Visual RAG, that synergically combines the MLLMs capability to learn from the context, with a retrieval mechanism. The crux of this approach is to ensure to augment the MLLM knowledge by selecting only the most relevant demonstrating examples for the query, pushing it to learn by analogy. In this way, relying on the new information provided dynamically during inference time, the resulting system is not limited to the knowledge extracted from the training data, but can be updated rapidly and easily without fine-tuning. Furthermore, this greatly reduces the computational costs for improving the model image classification performance, and augments the model knowledge to new visual domains and tasks it was not trained for. Extensive experiments on eight different datasets in the state of the art spanning several domains and image classification tasks show that the proposed Visual RAG, compared to the most recent state of the art (i.e., many-shot ICL), is able to obtain an accuracy that is very close or even higher (approx. +2% improvement on average) while using a much smaller set of demonstrating examples (approx. only 23% on average). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2501_10834 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Visual RAG: Expanding MLLM visual knowledge without fine-tuning Bonomo, Mirco Bianco, Simone Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning Multimodal Large Language Models (MLLMs) have achieved notable performance in computer vision tasks that require reasoning across visual and textual modalities, yet their capabilities are limited to their pre-trained data, requiring extensive fine-tuning for updates. Recent researches have explored the use of In-Context Learning (ICL) to overcome these challenges by providing a set of demonstrating examples as context to augment MLLMs performance in several tasks, showing that many-shot ICL leads to substantial improvements compared to few-shot ICL. However, the reliance on numerous demonstrating examples and the limited MLLMs context windows presents significant obstacles. This paper aims to address these challenges by introducing a novel approach, Visual RAG, that synergically combines the MLLMs capability to learn from the context, with a retrieval mechanism. The crux of this approach is to ensure to augment the MLLM knowledge by selecting only the most relevant demonstrating examples for the query, pushing it to learn by analogy. In this way, relying on the new information provided dynamically during inference time, the resulting system is not limited to the knowledge extracted from the training data, but can be updated rapidly and easily without fine-tuning. Furthermore, this greatly reduces the computational costs for improving the model image classification performance, and augments the model knowledge to new visual domains and tasks it was not trained for. Extensive experiments on eight different datasets in the state of the art spanning several domains and image classification tasks show that the proposed Visual RAG, compared to the most recent state of the art (i.e., many-shot ICL), is able to obtain an accuracy that is very close or even higher (approx. +2% improvement on average) while using a much smaller set of demonstrating examples (approx. only 23% on average). |
| title | Visual RAG: Expanding MLLM visual knowledge without fine-tuning |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2501.10834 |