Research map / Image generation / Text-to-image diffusion
Improving Controllable Generation: Faster Training and Better Performance via x0-Supervision
Amadou S. Sangare, Adrien Maglo, Mohamed Chaouch, Bertrand Luvison
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 |
|---|---|---|---|
| supervise with clean image x0 instead of epsilon | ControlNet · Semantic Seg (ADE20K) | mAUCC: 25.19 -> 31.52 | Improved |
| supervise with clean image x0 instead of epsilon | ControlNet · Pose (MS-COCO) | mAUCC: 35.86 -> 42.19 | Improved |
| supervise with x0 instead of epsilon | T2I-Adapter · Pose (MS-COCO) | mAUCC: 12.98 -> 21.45 | Improved |
| supervise with x0 instead of epsilon | T2I-Adapter · Pose (MS-COCO) | mAP pose: 29.50 -> 37.75 | Improved |
| supervise with x0 instead of u (flow matching) | OminiControl · Depth (MultiGen-20M) | mAUCC: 13.47 -> 13.74 | No clear effect |
| supervise with x0 instead of u (flow matching) | OminiControl · Pose (MS-COCO) | mAUCC: 34.54 -> 42.07 | Improved |
| supervise with x0 instead of u (flow matching) | OminiControl · Semantic Seg (ADE20K) | mIoU: 42.09 -> 45.60 | Improved |
| supervise with x0 instead of epsilon | GLIGEN · GLIGEN Box+Text (MS-COCO) | mAUCC: 8.28 -> 18.38 | Improved |
| supervise with x0 instead of epsilon | GLIGEN · GLIGEN Box+Text+Image (MS-COCO) | mAP box: 21.31 -> 20.76 | No clear effect |
| use v-supervision instead of epsilon (ablation) | ControlNet · Semantic Seg (ADE20K) | mIoU: 35.84 -> 31.11 | Hurt |
| use epsilon-supervision instead of u for flow matching (ablation) | OminiControl · Pose (MS-COCO) | mAP pose: 66.53 -> 54.66 | Hurt |
| reweight epsilon loss by inverse SNR (equivalent to x0-supervision) (ablation) | ControlNet · all tasks | convergence speed | Improved |
Closest papers by meaning
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.92)
- Simplifying Control Mechanism in Text-to-Image Diffusion Models (AAAI 2025 · similarity 0.91)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.90)
- Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance (NeurIPS 2024 · similarity 0.90)
- T2I-Adapter: Learning Adapters to Dig Out More Controllable Ability for Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.90)
- Diffusion Self-Distillation for Zero-Shot Customized Image Generation (CVPR 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.
