Research map / Image generation / Text-to-image diffusion
Dual Recursive Feedback on Generation and Appearance Latents for Pose-Robust Text-to-Image Diffusion
Jiwon Kim, Pureum Kim, SeonHwa Kim, Soobin Park, Eunju Cha, Kyong Hwan Jin
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 DRF (use DRF instead of Ctrl-X alone) | Ctrl-X · All structure types | successive rate: 0.9577 -> 0.9859 | Improved |
| use DRF instead of Ctrl-X | Ctrl-X · Pose structure | DINO-I: 0.7378 -> 0.7391 | No clear effect |
| use DRF instead of Ctrl-X | Ctrl-X · Canny structure | DINO-I: 0.7335 -> 0.7135 | Hurt |
| use DRF instead of Ctrl-X | Ctrl-X · Point cloud structure | DINO Self-Sim: 0.1205 -> 0.1114 | Improved |
| use DDIM 40 steps with DRF replaced by DPM-Solver++ 10 steps (ablation) | DRF with DDIM · Scheduler study | Time (s): 47.87 -> 15.56 | Improved |
| apply DRF for 10 steps instead of 20 steps (ablation) | DRF · Step ablation | inference time (s): 56.87 -> 36.09 | Mixed |
| apply DRF at all 50 steps instead of 20 steps (ablation) | DRF · Step ablation | inference time (s): 56.87 -> 135.07 | No clear effect |
| increase DRF iterations N from 1 to 3 (ablation) | DRF · Inference efficiency | inference latency (s): 29.69 -> 56.87 | Mixed |
| use SD 1.5 backbone instead of SDXL for DRF | DRF · Cross-backbone | GPU memory (GiB): 18.98 -> 5.87 | Improved |
| use SD 2.0 backbone instead of SDXL for DRF | DRF · Cross-backbone | DINO-I: 0.6957 -> 0.453 | Hurt |
| use exponential weighting of generation feedback instead of linear or uniform (ablation) | DRF · Weight schedule ablation | image quality (qualitative) | Improved |
| increase recursion depth K beyond 5 (ablation) | DRF · Hyper-parameter sweep | CLIP, Self-Sim, DINO-I | No clear effect |
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.
- FreeControl — FreeControl: Training-Free Spatial Control of Any Text-to-Image Diffusion Model with Any Condition (CVPR 2024) · better on 11 of 13 shared results
- Ctrl-X — Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance (NeurIPS 2024) · better on 11 of 13 shared results
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
- Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance (NeurIPS 2024 · similarity 0.89)
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.88)
- Referee Can Play: An Alternative Approach to Conditional Generation via Model Inversion (ICML 2024 · similarity 0.88)
- PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor (CVPR 2024 · similarity 0.87)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.87)
- David and Goliath: Small One-step Model Beats Large Diffusion with Score Post-training (ICML 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.
