# Few-Shot VQA with Frozen LLMs: A Tale of Two Approaches

arXiv: https://arxiv.org/abs/2403.11317 · HTML: https://mlsanity.com/paper/2403.11317/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 29 references that are in the map.
- First posted on arXiv 17 Mar 2024.

## Builds on (most cited first, 22 of 22)

- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [LAION-5B: An open large-scale dataset for training next generation image-text models](https://mlsanity.com/paper/2210.08402/) (2022)
- [Scaling Instruction-Finetuned Language Models](https://mlsanity.com/paper/2210.11416/) (2022)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [VQA: Visual Question Answering](https://mlsanity.com/paper/1505.00468/) (2015)
- [Billion-scale similarity search with GPUs](https://mlsanity.com/paper/1702.08734/) (2017)
- [What Makes Good In-Context Examples for GPT-3?](https://mlsanity.com/paper/2101.06804/) (2021)
- [PaLI: A Jointly-Scaled Multilingual Language-Image Model](https://mlsanity.com/paper/2209.06794/) (2022)
- [Multimodal Few-Shot Learning with Frozen Language Models](https://mlsanity.com/paper/2106.13884/) (2021)
- [SimVLM: Simple Visual Language Model Pretraining with Weak Supervision](https://mlsanity.com/paper/2108.10904/) (2021)
- [ClipCap: CLIP Prefix for Image Captioning](https://mlsanity.com/paper/2111.09734/) (2021)
- [Language Is Not All You Need: Aligning Perception with Language Models](https://mlsanity.com/paper/2302.14045/) (2023)
- [VL-Adapter: Parameter-Efficient Transfer Learning for Vision-and-Language Tasks](https://mlsanity.com/paper/2112.06825/) (2021)
- [An Empirical Study of GPT-3 for Few-Shot Knowledge-Based VQA](https://mlsanity.com/paper/2109.05014/) (2021)
- [Linearly Mapping from Image to Text Space](https://mlsanity.com/paper/2209.15162/) (2022)
- [Yin and Yang: Balancing and Answering Binary Visual Questions](https://mlsanity.com/paper/1511.05099/) (2015)
- [Retrieval Augmented Visual Question Answering with Outside Knowledge](https://mlsanity.com/paper/2210.03809/) (2022)
- [Plug-and-Play VQA: Zero-shot VQA by Conjoining Large Pretrained Models with Zero Training](https://mlsanity.com/paper/2210.08773/) (2022)
- [MAGMA -- Multimodal Augmentation of Generative Models through Adapter-based Finetuning](https://mlsanity.com/paper/2112.05253/) (2021)

## Built on by (most cited first, 4 of 4)

- [EmoLLM: Multimodal Emotional Understanding Meets Large Language Models](https://mlsanity.com/paper/2406.16442/) (2024)
- [VideoICL: Confidence-based Iterative In-context Learning for Out-of-Distribution Video Understanding](https://mlsanity.com/paper/2412.02186/) (2024)
- [FEALLM: Advancing Facial Emotion Analysis in Multimodal Large Language Models with Emotional Synergy and Reasoning](https://mlsanity.com/paper/2505.13419/) (2025)
- [Reasoning Like Experts: Leveraging Multimodal Large Language Models for Drawing-based Psychoanalysis](https://mlsanity.com/paper/2510.19451/) (2025)

## Abstract (arXiv)

Two approaches have emerged to input images into large language models (LLMs). The first is to caption images into natural language. The second is to map image feature embeddings into the domain of the LLM and pass the mapped embeddings directly to the LLM. The majority of recent few-shot multimodal work reports performance using architectures that employ variations of one of these two approaches. But they overlook an important comparison between them. We design a controlled and focused experiment to compare these two approaches to few-shot visual question answering (VQA) with LLMs. Our findings indicate that for Flan-T5 XL, a 3B parameter LLM, connecting visual embeddings directly to the LLM embedding space does not guarantee improved performance over using image captions. In the zero-shot regime, we find using textual image captions is better. In the few-shot regimes, how the in-context examples are selected determines which is better.