Research map / Vision-language and multimodal / Visual question answering
Fine-grained Late-interaction Multi-modal Retrieval for Retrieval Augmented Visual Question Answering
Weizhe Lin, Jinghong Chen, Jingbiao Mei, Alexandru Coca, Bill Byrne
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| upgrade retriever from DPR to FLMR (late interaction) | RA-VQA-v2 (T5-large) · OK-VQA | VQA Score: 51.45 -> 54.85 | Improved |
| add retrieval augmentation to BLIP 2 | BLIP 2 · OK-VQA | VQA Score: 55.44 -> 62.08 | Improved |
| remove late interaction (use DPR) (ablation) | RA-VQA-v2 (BLIP 2) · OK-VQA | VQA Score: 58.20 -> 62.08 | Improved |
| add region-of-interest features (ablation) | RA-VQA-v2 (BLIP 2) · OK-VQA | VQA Score: 60.41 -> 62.08 | Improved |
| add text-based vision (captions, OCR, objects) | FLMR · OK-VQA GS | PRRecall@5: 74.81 -> 85.99 | Improved |
| add feature-based vision (vision encoder) | FLMR · OK-VQA GS | PRRecall@5: 74.81 -> 85.08 | Improved |
| combine text-based and feature-based vision | FLMR · OK-VQA GS | PRRecall@5: 85.99 -> 87.02 | Improved |
| add 9 ROIs to the retriever | FLMR · OK-VQA GS | PRRecall@5: 87.02 -> 89.32 | Improved |
| add 9 ROIs to the retriever | DPR · OK-VQA GS | PRRecall@5: 83.43 -> 82.90 | Hurt |
| remove vision-language alignment pre-training (ablation) | FLMR · OK-VQA GS | PRRecall@5: 85.71 -> 89.32 | Improved |
| increase retrieved documents K from 5 to 50 | DPR + T5-large · OK-VQA | VQA Score: 51.5 -> 52.1 | Mixed |
| increase retrieved documents K from 5 to 50 | FLMR + T5-large · OK-VQA | VQA Score: 54.9 -> 55.4 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.
- PALI (17B) — PaLI: A Jointly-Scaled Multilingual Language-Image Model (ICLR 2023) · better on 1 of 2 shared results
- PaLM-E (562B) — PaLM-E: An Embodied Multimodal Language Model (ICML 2023) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VQA4CIR: Boosting Composed Image Retrieval with Visual Question Answering (AAAI 2025 · similarity 0.92)
- Retrieval-Augmented Visual Question Answering via Built-in Autoregressive Search Engines (AAAI 2025 · similarity 0.90)
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.90)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.90)
- FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering (NeurIPS 2025 · similarity 0.90)
- Knowledge-based Visual Question Answer with Multimodal Processing, Retrieval and Filtering (NeurIPS 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
