Research map / Image generation / Text-to-image diffusion
SmartControl: Enhancing ControlNet for Handling Rough Visual Conditions
Xiaoyu Liu, Yuxiang Wei, Ming Liu, Xianhui Lin, Peiran Ren, xuansong xie, Wangmeng Zuo
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 |
|---|---|---|---|
| predict a spatially local control scale map instead of a fixed scale (ablation) | SmartControl · Depth rough condition | CLIP Score: 0.268 -> 0.274 | Improved |
| predict local scale map instead of one global scale (ablation) | SmartControl · Depth rough condition | CLIP Score: 0.272 -> 0.274 | Improved |
| train control scale predictor instead of fine-tuning ControlNet (ablation) | SmartControl · Depth rough condition | CLIP Score: 0.248 -> 0.274 | Improved |
| train control scale predictor instead of an adapter (ablation) | SmartControl · Depth rough condition | Self-similarity distance: 0.192 -> 0.128 | Improved |
| use convolution predictor instead of cross-attention (ablation) | SmartControl · Depth rough condition | CLIP Score / time: CLIP 0.272 -> 0.274; time 7.69 -> 7.36 | Improved |
| increase training set size from 500 to 2000 images (ablation) | SmartControl · Depth rough condition | Class Confidence: 0.740 -> 0.742 | Improved |
| add conflict/background regularization loss L_c (ablation) | SmartControl · Qualitative | layout control and artifacts | Improved |
| SmartControl versus ControlNet alpha=1.0 on structure preservation | SmartControl · Depth rough condition | Self-similarity distance: 0.100 -> 0.128 | Hurt |
| SmartControl versus fixed-scale ControlNet on canny Class Confidence | SmartControl · Canny rough condition | Class Confidence: 0.736 -> 0.721 | Hurt |
Closest papers by meaning
- FreeControl: Training-Free Spatial Control of Any Text-to-Image Diffusion Model with Any Condition (CVPR 2024 · similarity 0.87)
- VersaGen: Unleashing Versatile Visual Control for Text-to-Image Synthesis (AAAI 2025 · similarity 0.86)
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.86)
- Focus-N-Fix: Region-Aware Fine-Tuning for Text-to-Image Generation (CVPR 2025 · similarity 0.85)
- FlexControl: Computation-Aware Conditional Control with Differentiable Router for Text-to-Image Generation (ICML 2025 · similarity 0.85)
- Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance (NeurIPS 2024 · similarity 0.85)
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.
