Research map / Vision-language and multimodal / Visual question answering
Unifying Vision-and-Language Tasks via Text Generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| generate answers instead of using a discriminative answer classifier | VL-T5 on VQA · VQA Karpathy-test out-of-domain | accuracy: 7.1 -> 13.1 | Helped |
| generate answers instead of using a discriminative answer classifier | VL-BART on VQA · VQA Karpathy-test out-of-domain | accuracy: 7.0 -> 13.2 | Helped |
| use one vqa prefix for VQA and GQA instead of dataset-specific prefixes (ablation) | VL-T5 · VQA Karpathy-test | accuracy: 67.9→69.3 | Helped |
| use one vqa prefix for VQA and GQA instead of dataset-specific prefixes (ablation) | VL-T5 · GQA test-dev | accuracy: 60.0→60.2 | Helped |
| add vision-and-language pretraining (ablation) | VL-T5 · RefCOCOg testᵈ | Acc: 62.9 -> 71.3 | Helped |
| add Stage 1 generic-domain pretraining (ablation) | VL-T5 · VCR val Q→AR | Acc: 52.5 -> 54.7 | Helped |
| add Stage 2 in-domain VCR pretraining after Stage 1 (ablation) | VL-T5 · VCR val Q→AR | Acc: 54.7 -> 57.5 | Helped |
| add object tags during finetuning (ablation) | VL-T5 · COCO Caption Karpathy test | CIDEr: 116.1 -> 116.5 | Helped |
| add object tags during finetuning (ablation) | VL-BART · COCO Caption Karpathy test | CIDEr: +2.5 CIDEr | Helped |
| add vision-and-language pretraining (ablation) | VL-T5 · Multi30K test 2018 | BLEU: 39.5 -> 38.6 | Hurt |
| finetune one shared VL-T5 on all seven tasks rather than one model per task (ablation) | VL-T5 · VQA Karpathy test | Acc: 67.9 -> 67.2 | Hurt |
| finetune one shared VL-T5 on all seven tasks rather than one model per task (ablation) | VL-T5 · COCO Caption Karpathy test | CIDEr: 116.1 -> 110.8 | Hurt |
Closest papers by meaning
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.90)
- Open-ended VQA benchmarking of Vision-Language models by exploiting Classification datasets and their semantic hierarchy (ICLR 2024 · similarity 0.90)
- Unified Generative and Discriminative Training for Multi-modal Large Language Models (NeurIPS 2024 · similarity 0.89)
- Marten: Visual Question Answering with Mask Generation for Multi-modal Document Understanding (CVPR 2025 · similarity 0.89)
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 2024 · similarity 0.89)
- VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks (NeurIPS 2024 · similarity 0.89)
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.