Research map / Vision-language and multimodal / Human and embodied modeling
PaLI: A Jointly-Scaled Multilingual Language-Image Model
Xi Chen, Xiao Wang, Soravit Changpinyo, AJ Piergiovanni, Piotr Padlewski, Daniel Salz, Sebastian Goodman, Adam Grycner, Basil Mustafa, Lucas Beyer, Alexander Kolesnikov, Joan Puigcerver and 17 more
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 vision encoder from ViT-G to ViT-e | PaLI · COCO Captions and OKVQA | performance per 1B added parameters: +2.2 per 1B more ViT params; +0.4 per 1B more mT5 params | Improved |
| Remove split-captioning objective (ablation) | PaLI-3B · COCO Captions Karpathy-test | CIDEr: 141.4 -> 140.4 (-1.0) | Hurt |
| Remove split-captioning objective (ablation) | PaLI-3B · XM-3600 EN | CIDEr: 93.8 -> 87.5 (-6.3) | Hurt |
| Remove CC3M-35L captioning objective (ablation) | PaLI-3B · XM-3600 | CIDEr: 93.8 -> 94.9 (+1.1) EN; 42.5 -> 39.9 (-2.6) 6L | Mixed |
| Remove OCR objective (ablation) | PaLI-3B · TextVQA | score: 41.6 -> 39.9 (-1.7) | Hurt |
| Remove OCR objective (ablation) | PaLI-3B · COCO Captions | CIDEr: 141.4 -> 142.3 (+0.9) | Improved |
| Remove VQA objective (ablation) | PaLI-3B · TextVQA | score: 41.6 -> 40.0 (-1.6) | Hurt |
| Remove VQG objective (ablation) | PaLI-3B · VQG zero-shot | score: 96.7 -> 17.9 (-78.8) | Hurt |
| Remove object-aware QA and object detection objectives (ablation) | PaLI-3B · XM-3600 EN | CIDEr: 93.8 -> 90.9 (-2.9) | Hurt |
| Initialize from scratch instead of pretrained unimodal checkpoints (ablation) | PaLI-3B · COCO Captions Karpathy-test | CIDEr: 141.4 -> 72.8 | Hurt |
| Freeze ViT during pretraining instead of fine-tuning it (ablation) | PaLI-3B · COCO Captions Karpathy-test | CIDEr: 138.8 -> 139.3 | Improved |
| Include multilingual WebLI data rather than only WebLI English (ablation) | 1.5B PaLI · XM-3600 6L | CIDEr: 8.2 -> 30.0 | Improved |
Later papers that compared with this method
9 later accepted papers report PaLI-17B as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Reverse Region-to-Entity Annotation for Pixel-Level Visual Entity Linking (AAAI 2025 · better on 1 of 1 shared result)
- Prophet: Prompting Large Language Models With Complementary Answer Heuristics for Knowledge-Based Visual Question Answering (TPAMI 2025 · better on 1 of 1 shared result)
- Grounding Language Models for Visual Entity Recognition (ECCV 2024 · better on 5 of 5 shared results)
- BRAVE: Broadening the visual encoding of vision-language models (ECCV 2024 · better on 2 of 4 shared results)
- A Generative Approach for Wikipedia-Scale Visual Entity Recognition (CVPR 2024 · better on 2 of 2 shared results)
- DocFormerv2: Local Features for Document Understanding (AAAI 2024 · better on 0 of 1 shared result)
- mPLUG-Owl2: Revolutionizing Multi-modal Large Language Model with Modality Collaboration (CVPR 2024 · better on 0 of 1 shared result)
- Fine-grained Late-interaction Multi-modal Retrieval for Retrieval Augmented Visual Question Answering (NeurIPS 2023 · better on 1 of 2 shared results)
1 more not listed.
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.
- BEiT-3 — BEiT: BERT Pre-Training of Image Transformers (ICLR 2022) · better on 2 of 2 shared results
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
- DiM$^3$: Bridging Multilingual and Multimodal Models via Direction- and Magnitude-Aware Merging (NeurIPS 2026 · similarity 0.90)
- On Scaling Up a Multilingual Vision and Language Model (CVPR 2024 · similarity 0.90)
- SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models (ICML 2024 · similarity 0.90)
- LRM-LLaVA: Overcoming the Modality Gap of Multilingual Large Language-Vision Model for Low-Resource Languages (AAAI 2025 · similarity 0.90)
- ROSE: Rotate Your Large Language Model to See (CVPR 2026 · similarity 0.90)
- The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models (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.
