Research map / Image generation / Text-to-image diffusion
ERNIE-ViLG 2.0: Improving Text-to-Image Diffusion Model With Knowledge-Enhanced Mixture-of-Denoising-Experts
Zhida Feng, Zhenyu Zhang, Xintong Yu, Yewei Fang, Lanxin Li, Xuyi Chen, Yuxiang Lu, Jiaxiang Liu, Weichong Yin, Shikun Feng, Yu Sun, Li Chen and 3 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 |
|---|---|---|---|
| Add textual and visual knowledge enhancement (ablation) | ERNIE-ViLG 2.0 · text-to-image generation | image fidelity, image-text alignment, and convergence speed | Improved |
| Use object knowledge without synthetic descriptions (ablation) | ERNIE-ViLG 2.0 · text-to-image generation | overall performance | No clear effect |
| Add synthetic descriptions alongside object knowledge (ablation) | ERNIE-ViLG 2.0 · text-to-image generation | usefulness of visual knowledge | Improved |
| Add textual knowledge (ablation) | ERNIE-ViLG 2.0 · fine-grained semantic control | CLIP Score: +0.0051 CLIP Score on Counterfactual; +0.0041 on Color | Improved |
| Add visual knowledge (ablation) | ERNIE-ViLG 2.0 · fine-grained image-text alignment | CLIP Score: +0.0080 CLIP Score on Counterfactual; +0.0047 on Counting | Improved |
| Combine textual and visual knowledge (ablation) | ERNIE-ViLG 2.0 · ViLG-300 Complex | CLIP Score: +0.0074 CLIP Score | Improved |
| Increase denoising experts from 1 to 10 (ablation) | ERNIE-ViLG 2.0 · MS-COCO 256×256 zero-shot, without reranking | FID-30k: 8.07 -> 7.23 | Improved |
| Use multiple timestep-specific experts rather than one expert, at matched training samples (ablation) | ERNIE-ViLG 2.0 · text-to-image generation | overall performance | Improved |
| Increase the number of denoising experts (ablation) | ERNIE-ViLG 2.0 · text-to-image generation | image quality | Improved |
| Use 10 denoising experts instead of one (ablation) | ERNIE-ViLG 2.0 · MS-COCO 256×256 zero-shot | FID-30k: 8.07 -> 7.23 without reranking; 7.62 -> 6.75 with reranking | Improved |
Later papers that compared with this method
2 later accepted papers report ERNIE-ViLG 2.0 as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Bridge Diffusion Model: Bridge Chinese Text-to-Image Diffusion Model with English Communities (AAAI 2025 · better on 0 of 1 shared result)
- RAPHAEL: Text-to-Image Generation via Large Mixture of Diffusion Paths (NeurIPS 2023 · 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
- Diffusion-based Blind Text Image Super-Resolution (CVPR 2024 · similarity 0.88)
- RealCompo: Balancing Realism and Compositionality Improves Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.88)
- AltDiffusion: A Multilingual Text-to-Image Diffusion Model (AAAI 2024 · similarity 0.88)
- Image Regeneration: Evaluating Text-to-Image Model via Generating Identical Image with Multimodal Large Language Models (AAAI 2025 · similarity 0.88)
- Towards Understanding the Working Mechanism of Text-to-Image Diffusion Model (NeurIPS 2024 · similarity 0.88)
- Enhancing Text-to-Image Diffusion Transformer via Split-Text Conditioning (NeurIPS 2025 · similarity 0.87)
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.
