Research map / Image generation / Autoregressive and latent generation
RelaCtrl: Relevance-Guided Efficient Control for Diffusion Transformers
Ke Cao, Jing Wang, Ao Ma, Jiasong Feng, Xuanhua He, Run Ling, Haowei Liu, Jian Lu, Wei Feng, Haozhe Wang, Hongjuan Pei, Yihua Shao and 2 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 |
|---|---|---|---|
| use RelaCtrl lightweight relevance-guided control instead of PixArt-delta copy | PixArt-alpha ControlNet · Canny (COCO val) | FID: 21.38 -> 20.34 | Improved |
| use RelaCtrl instead of PixArt-delta | PixArt-alpha ControlNet · Segmentation (COCO val) | mIoU: 0.379 -> 0.405 | Improved |
| use RelaCtrl instead of full ControlNet copy | PixArt-alpha · parameters | extra parameters (M): +294.34 -> +45.15 | Improved |
| use RelaCtrl instead of full ControlNet copy | PixArt-alpha · complexity | extra GFLOPs: +270.57 -> +46.71 | Improved |
| place 13 control blocks by relevance rank instead of first 13 blocks (ablation) | ControlNet · Canny | FID: 21.38 -> 20.31 | Improved |
| reduce relevance-ranked control blocks from 13 to 11 (ablation) | RelaCtrl · Canny | FID: 20.31 -> 21.28 | No clear effect |
| reduce relevance-ranked control blocks from 11 to 10 (ablation) | RelaCtrl · Canny | FID: 21.28 -> 22.24 | Hurt |
| add RGLC block (TDSM) instead of the original copy block (ablation) | RelaCtrl · Canny | FID: 21.28 -> 20.34 | Improved |
| add relevance-based TDSM channel division (Prior 2) (ablation) | RelaCtrl · Canny | FID: 22.47 -> 20.34 | Improved |
| add relevance-based TDSM channel division (Prior 2) (ablation) | RelaCtrl · Canny | HDD: 97.30 -> 94.04 | Improved |
Closest papers by meaning
- Scaling Diffusion Transformers Efficiently via $\mu$P (NeurIPS 2025 · similarity 0.87)
- FlexControl: Computation-Aware Conditional Control with Differentiable Router for Text-to-Image Generation (ICML 2025 · similarity 0.87)
- FasterDiT: Towards Faster Diffusion Transformers Training without Architecture Modification (NeurIPS 2024 · similarity 0.87)
- PTQ4DiT: Post-training Quantization for Diffusion Transformers (NeurIPS 2024 · similarity 0.86)
- Accelerating Diffusion Transformer via Increment-Calibrated Caching with Channel-Aware Singular Value Decomposition (CVPR 2025 · similarity 0.86)
- VD3D: Taming Large Video Diffusion Transformers for 3D Camera Control (ICLR 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.
