Research map / Image generation / Editing and style transfer
CompleteMe: Reference-based Human Image Completion
Yu-Ju Tsai, Brian Price, Qing Liu, Luis Figueroa, Daniil Pakhomov, Zhihong Ding, Scott Cohen, Ming-Hsuan Yang
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 |
|---|---|---|---|
| replace CLIP encoder with Reference U-Net encoder for reference features (ablation) | CompleteMe · benchmark | DreamSim: 0.0457 -> 0.0437 | Improved |
| replace DINOv2 encoder with Reference U-Net encoder (ablation) | CompleteMe · benchmark | DreamSim: 0.0639 -> 0.0437 | Improved |
| add Region-focused Attention block to U-Net encoder (ablation) | CompleteMe · benchmark | DreamSim: 0.0437 -> 0.0419 | Improved |
| add Region-focused Attention block to U-Net encoder (ablation) | CompleteMe · benchmark | CLIP-I: 97.05 -> 97.18 | Improved |
| train the Reference U-Net instead of freezing it (ablation) | CompleteMe · benchmark | DreamSim: 0.0444 -> 0.0419 | Improved |
| add reference masks to RFA (with frozen U-Net and prompt) (ablation) | CompleteMe · benchmark | DreamSim: 0.0521 -> 0.0444 | Improved |
| add text prompt input to Complete U-Net (frozen U-Net, no mask) (ablation) | CompleteMe · benchmark | DreamSim: 0.0513 -> 0.0521 | No clear effect |
| use 50% random mask ratio instead of 100% random masks (ablation) | CompleteMe · benchmark | DreamSim: 0.0495 -> 0.0419 | Improved |
| use 50% random mask ratio instead of 0% (human shape only) (ablation) | CompleteMe · benchmark | CLIP-I: 97.09 -> 97.18 | No clear effect |
| CompleteMe vs MimicBrush | MimicBrush · benchmark | DreamSim: 0.0651 -> 0.0419 | Improved |
| CompleteMe vs LeftRefill on PSNR | LeftRefill · benchmark | PSNR: 28.8657 -> 28.7020 | Hurt |
| user study: CompleteMe vs LeftRefill on quality | LeftRefill · benchmark user study | Quality preference: 14.97% / 85.03% | Improved |
Later papers that compared with this method
1 later accepted paper reports CompleteMe as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- UNet-Twice: A Simple Structured Reference-based Inpainting Framework (ECCV 2026 · better on 2 of 2 shared results)
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.
- LeftRefill — LeftRefill: Filling Right Canvas based on Left Reference through Generalized Text-to-Image Diffusion Model (CVPR 2024) · better on 3 of 4 shared results
- BrushNet — BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion (ECCV 2024) · better on 4 of 4 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
- PHAC: Promptable Human Amodal Completion (CVPR 2026 · similarity 0.87)
- Large Occluded Human Image Completion via Image-Prior Cooperating (AAAI 2024 · similarity 0.87)
- Prior Based Human Completion (CVPR 2021 · similarity 0.87)
- GeoComplete: Geometry-Aware Diffusion for Reference-Driven Image Completion (NeurIPS 2025 · similarity 0.84)
- HumanRef: Single Image to 3D Human Generation via Reference-Guided Diffusion (CVPR 2024 · similarity 0.84)
- MaGIC: Multi-modality Guided Image Completion (ICLR 2024 · similarity 0.84)
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.
