Research map / Image generation / Text-to-image diffusion
HumanRefiner: Benchmarking Abnormal Human Generation and Refining with Coarse-to-fine Pose-Reversible Guidance
Guian Fang, Wenbiao Yan, Yuanfan Guo, Jianhua Han, Zutao Jiang, Hang Xu, Shengcai Liao, Xiaodan Liang
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 negative prompt learned from AbHuman annotations (ablation) | SDXL · AbHuman test (LAION) | Abnormal Score: 0.659 -> 0.614 | Improved |
| add abnormal scorer guidance (ablation) | SDXL with negative prompt · AbHuman test (LAION) | Abnormal Score: 0.614 -> 0.605 | Improved |
| add detector-guided inpainting (ablation) | HumanRefiner · AbHuman test (LAION) | Abnormal Score: 0.605 -> 0.601 | Improved |
| add pose-guided coarse refinement (ablation) | HumanRefiner · AbHuman test (LAION) | Abnormal Score: 0.601 -> 0.590 | Improved |
| use HumanRefiner instead of SDXL | SDXL · AbHuman hard test (HumanArt) | FID: 20.796 -> 9.145 | Improved |
| use HumanRefiner instead of SDXL in human evaluation of limb quality | SDXL · Human evaluation, HumanArt prompts | preference percentage: 11.20 -> 32.3 | Improved |
| use HumanRefiner instead of DALL-E 3 in human evaluation of limb quality | DALL-E 3 · Human evaluation, HumanArt prompts | preference percentage: 36.03 -> 51.42 | Improved |
| use HumanRefiner instead of Pose-ControlNet and Pose-T2I-Adapter on SDXL images | Pose-T2I-Adapter · AbHuman test (LAION) | Abnormal Score: 0.624 -> 0.590 | Improved |
| use YOLOv8x instead of YOLOv8n for the abnormal detector (ablation) | Abnormal detector · AbHuman test | mAP50: 0.333 -> 0.426 (mAP50); 0.282 -> 0.235 (mAP50-95) | Mixed |
Closest papers by meaning
- HyperHuman: Hyper-Realistic Human Generation with Latent Structural Diffusion (ICLR 2024 · similarity 0.88)
- MultiHuman-Testbench: Benchmarking Image Generation for Multiple Humans (NeurIPS 2025 · similarity 0.87)
- HumanRef: Single Image to 3D Human Generation via Reference-Guided Diffusion (CVPR 2024 · similarity 0.87)
- Is this Generated Person Existed in Real-world? Fine-grained Detecting and Calibrating Abnormal Human-body (CVPR 2025 · similarity 0.86)
- HumanVid: Demystifying Training Data for Camera-controllable Human Image Animation (NeurIPS 2024 · similarity 0.85)
- UniMC: Taming Diffusion Transformer for Unified Keypoint-Guided Multi-Class Image Generation (ICML 2025 · 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.
