Research map / Vision-language and multimodal / Human and embodied modeling
UNITER: UNiversal Image-TExt Representation Learning
Yen-Chun Chen, Linjie Li, Licheng Yu, Ahmed El Kholy Faisal Ahmed, Zhe Gan, Yu Cheng, Jingjing Liu
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 |
|---|---|---|---|
| add MLM/ITM/MRM pre-training tasks jointly instead of no pre-training (ablation) | UNITER-base · Meta-Sum of VQA, NLVR2, Flickr30K, RefCOCO+ | Meta-Sum: 314.34 -> 400.93 | Improved |
| add ITM pre-training only (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 346.24 -> 385.29 | Improved |
| add MRFR alone on top of text-only MLM init (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 346.24 -> 344.66 | No clear effect |
| add ITM to MLM (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 386.10 -> 393.04 | Improved |
| use MRC-kl rather than MRC as masked region modeling variant (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 393.97 -> 397.09 | Improved |
| combine MRC-kl and MRFR (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 397.09 -> 399.97 | Improved |
| add word-region alignment (WRA) pre-training (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 399.97 -> 400.93 | Improved |
| use conditional masking instead of joint random masking (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 396.51 -> 399.97 | Improved |
| pre-train on out-of-domain data only instead of in-domain (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 400.93 -> 396.91 | Hurt |
| add out-of-domain data to in-domain pre-training data (ablation) | UNITER-base · Meta-Sum | Meta-Sum: 400.93 -> 405.24 | Improved |
| add WRA pre-training to UNITER-large (ablation) | UNITER-large · Flickr30K zero-shot text retrieval | R@1: 77.50 -> 83.60 | Improved |
| add second-stage pre-training on VCR data (ablation) | UNITER-large · VCR | Q->AR accuracy: 53.52 -> 55.6 | Improved |
Later papers that compared with this method
5 later accepted papers report UNITER Base as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- The Abduction of Sherlock Holmes: A Dataset for Visual Abductive Reasoning (ECCV 2022 · better on 6 of 6 shared results)
- Unifying Vision-and-Language Tasks via Text Generation (ICML 2021 · better on 1 of 2 shared results)
- ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision (ICML 2021 · better on 0 of 1 shared result)
- COOKIE: Contrastive Cross-Modal Knowledge Sharing Pre-Training for Vision-Language Representation (ICCV 2021 · better on 1 of 1 shared result)
- Large-Scale Adversarial Training for Vision-and-Language Representation Learning (NeurIPS 2020 · better on 4 of 4 shared results)
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.
- VL-BERT (Large) — VL-BERT: Pre-training of Generic Visual-Linguistic Representations (ICLR 2020) · 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
- VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks (ICLR 2025 · similarity 0.91)
- Text as Partial Constraint: Core–Residual Alignment for Robust Vision–Language Learning (NeurIPS 2026 · similarity 0.89)
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models (NeurIPS 2025 · similarity 0.89)
- DH-Set: Improving Vision-Language Alignment with Diverse and Hybrid Set-Embeddings Learning (CVPR 2025 · similarity 0.89)
- Analyzing Fine-Grained Alignment and Enhancing Vision Understanding in Multimodal Language Models (NeurIPS 2025 · similarity 0.89)
- MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer (ICLR 2026 · 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.
