Research map / Vision-language and multimodal / Human and embodied modeling
Unified Vision-Language Pre-Training for Image Captioning and VQA
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 |
|---|---|---|---|
| Apply Unified VLP rather than no VLP pre-training (ablation) | UniLM-initialized model · Flickr30k | CIDEr: 56.8 -> 67.4 | Helped |
| Apply Unified VLP rather than no VLP pre-training (ablation) | UniLM-initialized model · COCO Captions | CIDEr: 114.3 -> 116.9 | Helped |
| Use seq2seq pre-training only rather than no VLP pre-training (ablation) | UniLM-initialized model · COCO Captions | CIDEr: 114.3 -> 117.7 | Helped |
| Use bidirectional pre-training only rather than seq2seq pre-training only (ablation) | UniLM-initialized model · VQA 2.0 | Overall accuracy: 70.2 -> 71.3 | Helped |
| Use seq2seq pre-training only rather than bidirectional pre-training only (ablation) | UniLM-initialized model · Flickr30k | CIDEr: 63.3 -> 68.5 | Helped |
| Combine seq2seq and bidirectional objectives in Unified VLP rather than use seq2seq only (ablation) | UniLM-initialized model · VQA 2.0 | Overall accuracy: 70.2 -> 70.7; 71.3 -> 70.7 | Mixed |
| Predict region class labels as a pretext rather than feed class probabilities as input (ablation) | Vision-language pre-training · Conceptual Captions val | CIDEr: 67.0 -> 62.2 | Hurt |
| Initialize VLP from UniLM rather than train from scratch (ablation) | Vision-language pre-training · Conceptual Captions val | CIDEr: 63.8 -> 67.0 | Helped |
Closest papers by meaning
- Probing Visual Language Priors in VLMs (ICML 2025 · similarity 0.91)
- Vision‑Language‑Vision Auto‑Encoder: Scalable Knowledge Distillation from Diffusion Models (NeurIPS 2025 · similarity 0.91)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.89)
- Unveiling Encoder-Free Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Probabilistic Language-Image Pre-Training (ICLR 2025 · similarity 0.88)
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 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.