Research map / Vision-language and multimodal / CLIP and zero-shot
Oscar: Object-Semantics Aligned Pre-training for Vision-Language Tasks
Xiujun Li, Xi Yin, Chunyuan Li, Pengchuan Zhang, Xiaowei Hu, Lei Zhang, Lijuan Wang, Houdong Hu, Li Dong, Furu Wei, Yejin Choi, Jianfeng Gao
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 Visual Genome object tags as anchor points (ablation) | Oscar_B · VQA dev (pre-training ablation) | accuracy: 70.93 -> 71.70 | Improved |
| add Visual Genome object tags (ablation) | Oscar_B · COCO 1K text retrieval | R@1: 84.4 -> 88.4 | Improved |
| add Visual Genome object tags (ablation) | Oscar_B · COCO 1K image retrieval | R@1: 73.1 -> 75.7 | Improved |
| add Visual Genome object tags (ablation) | Oscar_B · COCO captioning | CIDEr: 115.6 -> 123.4 | Improved |
| add Visual Genome object tags (ablation) | Oscar_B · COCO captioning | BLEU@4: 34.5 -> 36.4 | Improved |
| add Open Images object tags (ablation) | Oscar_B · VQA dev (pre-training ablation) | accuracy: 70.93 -> 71.15 | Improved |
| add Open Images object tags (ablation) | Oscar_B · COCO 1K image retrieval | R@1: 73.1 -> 72.9 | Hurt |
| add Open Images object tags (ablation) | Oscar_B · COCO captioning | CIDEr: 115.6 -> 119.5 | Improved |
| use VG-detector tags instead of OI-detector tags (ablation) | Oscar_B · COCO captioning | CIDEr: 119.5 -> 123.4 | Improved |
| add object tags during fine-tuning (learning curves) (ablation) | Oscar_B · VQA and retrieval fine-tuning | convergence speed | Improved |
| add object tags as features via attention (w-v vs full attention) (ablation) | Oscar_B (no pretraining) · COCO 1K retrieval | R@1 | Improved |
| scale Oscar from BERT base to BERT large | Oscar · COCO captioning | CIDEr: 137.6 -> 140.0 | Improved |
Closest papers by meaning
- Tag2Text: Guiding Vision-Language Model via Image Tagging (ICLR 2024 · similarity 0.89)
- Analyzing Fine-Grained Alignment and Enhancing Vision Understanding in Multimodal Language Models (NeurIPS 2025 · similarity 0.89)
- DINOv2 Meets Text: A Unified Framework for Image- and Pixel-Level Vision-Language Alignment (CVPR 2025 · similarity 0.89)
- Text as Partial Constraint: Core–Residual Alignment for Robust Vision–Language Learning (NeurIPS 2026 · similarity 0.88)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.88)
- Visual-Text Cross Alignment: Refining the Similarity Score in Vision-Language Models (ICML 2024 · similarity 0.88)
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.
