Research map / Vision-language and multimodal / Human and embodied modeling
UNIFIED-IO: A Unified Model for Vision, Language, and Multi-modal Tasks
Jiasen Lu, Christopher Clark, Rowan Zellers, Roozbeh Mottaghi, Aniruddha Kembhavi
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 |
|---|---|---|---|
| scale model from SMALL to XL (71M to 2.9B params) | Unified-IO · GRIT test | average accuracy: 45.4 to 64.3 | Improved |
| remove image synthesis task group during multi-task training (ablation) | Unified-IO LARGE · GRIT keypoint (ablation split) | keypoint accuracy: 69.2 to 18.3 | Hurt |
| remove captioning task group (ablation) | Unified-IO LARGE · GRIT VQA and localization | accuracy: VQA 65.7 to 68.0 (improves when removed) | No clear effect |
| remove NLP task group during multi-task training (ablation) | Unified-IO LARGE · GRIT categorization | categorization accuracy: 50.3 to 56.1 when removed (removing helps) | Hurt |
| remove task groups generally (dense, sparse, classification, V&L, language modelling) (ablation) | Unified-IO LARGE · GRIT and MNLI | task accuracy | No clear effect |
| use pre-trained VQ-GAN image tokenizer without surface normal data | Unified-IO XL · GRIT surface normal | accuracy: 44.3 vs NLL-AngMF 49.6 | Hurt |
| 2-stage keypoint inference (locate person, then keypoints) | Unified-IO XL · GRIT keypoint | keypoint accuracy: 68.1 vs Mask R-CNN 70.8 | Hurt |
| use prompt paraphrases for referring expressions instead of training prompt (ablation) | Unified-IO · GRIT referring expression ablation | Refexp score: 78.9 to as low as 0.1 | Hurt |
Later papers that compared with this method
4 later accepted papers report UnifiedIO-XL as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Visual Grounding for Object Questions (CVPR 2026 · better on 6 of 8 shared results)
- GiT: Towards Generalist Vision Transformer through Universal Language Interface (ECCV 2024 · better on 0 of 1 shared result)
- Uni-Perceiver v2: A Generalist Model for Large-Scale Vision and Vision-Language Tasks (CVPR 2023 · better on 1 of 1 shared result)
- All in Tokens: Unifying Output Space of Visual Tasks via Soft Token (ICCV 2023 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- UViM — UViM: A Unified Modeling Approach for Vision with Learned Guiding Codes (NeurIPS 2022) · better on 1 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
- Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision Language Audio and Action (CVPR 2024 · similarity 0.88)
- Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: A Comprehensive Evaluation (ICLR 2026 · similarity 0.86)
- Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks (CVPR 2024 · similarity 0.86)
- UniCompress: Token Compression for Unified Vision-Language Understanding and Generation (CVPR 2026 · similarity 0.86)
- Aligning and Prompting Everything All at Once for Universal Visual Perception (CVPR 2024 · similarity 0.86)
- VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks (NeurIPS 2024 · similarity 0.86)
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.
