Research map / Image generation / Editing and style transfer
Multimodal Unsupervised Image-to-image Translation
Xun Huang, Ming-Yu Liu, Serge Belongie, Jan Kautz
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 image reconstruction loss (ablation) | MUNIT · edges->shoes | human preference over MUNIT (%): 6.0% -> 50.0% | Improved |
| add content reconstruction loss (ablation) | MUNIT · edges->handbags | human preference over MUNIT (%): 9.3% -> 50.0% | Improved |
| add style reconstruction loss (ablation) | MUNIT · edges->shoes | LPIPS diversity: 0.070 -> 0.109 | Improved |
| add style reconstruction loss (ablation) | MUNIT · edges->handbags | LPIPS diversity: 0.139 -> 0.175 | Improved |
| add style reconstruction loss (ablation) | MUNIT · edges->shoes | human preference over MUNIT (%): 28.6% -> 50.0% | Improved |
| inject noise vectors into CycleGAN | CycleGAN · edges->shoes | LPIPS diversity: 0.010 -> 0.016 | No clear effect |
| inject noise vectors into CycleGAN | CycleGAN · edges->shoes | human preference over MUNIT (%): 36.0% -> 29.5% | Hurt |
| inject noise vectors into CycleGAN | CycleGAN · edges->handbags | human preference over MUNIT (%): 40.8% -> 45.1% | Improved |
| use MUNIT instead of UNIT | UNIT · Animal image translation (average of 6 tasks) | CIS: 0.115 -> 1.039 | Improved |
| use MUNIT instead of CycleGAN | CycleGAN · Animal image translation (average of 6 tasks) | IS: 0.813 -> 1.050 | Improved |
| train MUNIT unsupervised instead of BicycleGAN with paired supervision | BicycleGAN · edges->shoes | human preference over MUNIT (%): 56.7% -> 50.0% | No clear effect |
| apply Instance Normalization to VGG features in the perceptual loss (ablation) | MUNIT perceptual loss · day/night image pairs | VGG feature distance domain invariance | Improved |
Closest papers by meaning
- Unsupervised Image-to-Image Translation Using Domain-Specific Variational Information Bound (NeurIPS 2018 · similarity 0.93)
- COCO-FUNIT: Few-Shot Unsupervised Image Translation with a Content Conditioned Style Encoder (ECCV 2020 · similarity 0.90)
- Towards Instance-Level Image-To-Image Translation (CVPR 2019 · similarity 0.90)
- Memory-Guided Unsupervised Image-to-Image Translation (CVPR 2021 · similarity 0.89)
- Unsupervised Image-to-Image Translation with Density Changing Regularization (NeurIPS 2022 · similarity 0.89)
- Towards Identifiable Unsupervised Domain Translation: A Diversified Distribution Matching Approach (ICLR 2024 · 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.
