Research map / Image generation / Editing and style transfer
GP-UNIT: Generative Prior for Versatile Unsupervised Image-to-Image Translation
Shuai Yang, Liming Jiang, Ziwei Liu, Chen Change Loy
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 generative prior distillation for the content encoder (ablation) | GP-UNIT · Ablation test set | FID: 16.11 -> 15.29 | Improved |
| add generative prior distillation for the content encoder (ablation) | GP-UNIT · Ablation user study | content consistency: 0.02 -> 0.83 | Improved |
| add dynamic skip connection (ablation) | GP-UNIT · Ablation test set | FID: 15.83 -> 15.29 | Improved |
| add dynamic skip connection (ablation) | GP-UNIT · Ablation user study | content consistency: 0.15 -> 0.83 | Improved |
| add generative prior distillation (ablation) | GP-UNIT · Ablation test set | Diversity: 0.55 -> 0.51 | Hurt |
| add dynamic skip connection (ablation) | GP-UNIT · Ablation test set | Diversity: 0.52 -> 0.51 | Hurt |
| use 4 labeled pairs of semi-supervision (ablation) | GP-UNIT · Bird<->Car layout consistency | accuracy: 0.160 -> 0.895 | Improved |
| increase labeled pairs from 4 to 40 (ablation) | GP-UNIT · Bird<->Car layout consistency | accuracy: 0.895 -> 0.960 | Improved |
| use position loss L_pos instead of L2 on label maps (40 labeled) (ablation) | GP-UNIT · Bird<->Car layout consistency | accuracy: 0.245 -> 0.960 | Improved |
| semi-supervision with labeled pairs (user study) (ablation) | GP-UNIT · Bird/Car user study | consistent results: 17.3% -> 69.0% | Improved |
| raise content control l from 0 to 1 (ablation) | GP-UNIT · Dog <-> Cat | style loss and content loss: style 0.124 -> 0.109; content 0.845 -> 1.192 | Mixed |
| raise content control l from 0 to 1 (ablation) | GP-UNIT · Male <-> Female | style loss: 0.146 -> 0.145 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports GP-UNIT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- One-for-All: Towards Universal Domain Translation With a Single StyleGAN (TPAMI 2025 · better on 4 of 4 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- StarGAN2 — StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation (CVPR 2018) · better on 5 of 6 shared results
- U-GAT-IT — U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation (ICLR 2020) · better on 2 of 2 shared results
- COCO-FUNIT — COCO-FUNIT: Few-Shot Unsupervised Image Translation with a Content Conditioned Style Encoder (ECCV 2020) · better on 2 of 2 shared results
- TraVeLGAN — TraVeLGAN: Image-To-Image Translation by Transformation Vector Learning (CVPR 2019) · better on 2 of 2 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
- Unsupervised Image-to-Image Translation With Generative Prior (CVPR 2022 · similarity 0.97)
- Domain-Scalable Unpaired Image Translation via Latent Space Anchoring (TPAMI 2023 · similarity 0.91)
- Unsupervised Image-to-Image Translation Networks (NeurIPS 2017 · similarity 0.91)
- TraVeLGAN: Image-To-Image Translation by Transformation Vector Learning (CVPR 2019 · similarity 0.90)
- DA-GAN: Instance-Level Image Translation by Deep Attention Generative Adversarial Networks (CVPR 2018 · similarity 0.90)
- Few-Shot Unsupervised Image-to-Image Translation (ICCV 2019 · similarity 0.89)
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.
