Research map / Image generation / Text-to-image diffusion
ELDiff: When Evidential Learning Meets Text-to-Image Diffusion
Qingtao Pan, Kai Ye, Zhihao Dou, Bing Ji, Shuo Li
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 pixel evidence loss on top of plain LDM finetuning (ablation) | SD v1.4 finetune · VISOR OA | OA: 0.3821 -> 0.5466 | Improved |
| add token conflict loss to pixel evidence loss (ablation) | SD v1.4 finetune · VISOR OA | OA: 0.5466 -> 0.5563 | Improved |
| add token conflict loss to pixel evidence loss (ablation) | SD v1.4 finetune · MultiGen MG2 | MG2 success rate: 0.9821 -> 0.9736 | Hurt |
| add token conflict loss to pixel evidence loss (ablation) | SD v1.4 finetune · MultiGen MG5 | MG5 success rate: 0.3361 -> 0.3530 | Improved |
| add pixel evidence loss on top of plain LDM finetuning (ablation) | SD v1.4 finetune · COCO instance val | FID: 24.57 -> 21.17 | Improved |
| finetune with plain LDM loss only (vs frozen SD v1.4) (ablation) | SD v1.4 · COCO instance val | FID: 22.06 -> 24.57 | Hurt |
| ELDiff finetuning instead of TokenCompose | SD v2.1 · VISOR OA | OA: 0.6010 -> 0.7116 | Improved |
| ELDiff finetuning instead of TokenCompose | SD v3.5 Medium · VISOR OA | OA: 0.8233 -> 0.8644 | Improved |
| plain LDM finetuning (no consistency loss) vs frozen | SD v3.5 Medium · VISOR OA | OA: 0.8084 -> 0.8010 | Hurt |
| ELDiff finetuning vs frozen SDXL | SDXL · T2I-CompBench Color | color attribute binding: 0.6734 -> 0.7768 | Improved |
| ELDiff finetuning vs base Qwen-Image | Qwen-Image · GenEval Overall | overall: 0.87 -> 0.90 | Improved |
| ELDiff finetuning vs base Qwen-Image | Qwen-Image · GenEval 2 Soft-TIFA GM | Soft-TIFA GM: 34.57 -> 37.94 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TokenCompose — TokenCompose: Text-to-Image Diffusion with Token-level Supervision (CVPR 2024) · better on 5 of 6 shared results
- InitNO — InitNO: Boosting Text-to-Image Diffusion Models via Initial Noise Optimization (CVPR 2024) · 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
- I Think, Therefore I Diffuse: Enabling Multimodal In-Context Reasoning in Diffusion Models (ICML 2025 · similarity 0.88)
- EmerDiff: Emerging Pixel-level Semantic Knowledge in Diffusion Models (ICLR 2024 · similarity 0.87)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
- TokenCompose: Text-to-Image Diffusion with Token-level Supervision (CVPR 2024 · similarity 0.87)
- A Cat Is A Cat (Not A Dog!): Unraveling Information Mix-ups in Text-to-Image Encoders through Causal Analysis and Embedding Optimization (NeurIPS 2024 · similarity 0.86)
- LightFair: Towards an Efficient Alternative for Fair T2I Diffusion via Debiasing Pre-trained Text Encoders (NeurIPS 2025 · similarity 0.86)
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.
