Research map / Image generation / Editing and style transfer
Guided Image-to-Image Translation With Bi-Directional Feature Transformation
Badour AlBahar, Jia-Bin Huang
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 bi-directional feature transformation instead of input concatenation (ablation) | guided pix2pix (ResNet) · NYU v2 depth upsampling x16 | RMSE (cm): 11.86 -> 9.01 | Improved |
| add bi-directional feature transformation instead of input concatenation (ablation) | guided pix2pix (ResNet) · DeepFashion pose transfer | FID: 42.330 -> 13.240 | Improved |
| make feature transformation bi-directional instead of uni-directional (ablation) | FT model · NYU v2 depth upsampling x16 | RMSE (cm): 11.41 -> 9.01 | Improved |
| make feature transformation bi-directional instead of uni-directional (ablation) | FT model · DeepFashion pose transfer | FID: 13.988 -> 13.240 | Improved |
| make feature transformation bi-directional instead of uni-directional (ablation) | FT model · Texture transfer, Shoes | FID: 119.588 -> 119.832 | No clear effect |
| use 4 FT layers instead of 1 in bFT (ablation) | bFT · NYU v2 depth upsampling x16 | RMSE (cm): 10.79 -> 9.01 | Improved |
| use 4 FT layers instead of 3 in bFT (ablation) | bFT · NYU v2 depth upsampling x16 | RMSE (cm): 8.82 -> 9.01 | Hurt |
| use 3 FT layers instead of 1 in uFT pose transfer (ablation) | uFT · DeepFashion pose transfer | FID: 59.678 -> 16.069 | Improved |
| use FT instead of AdaIN in bi-directional conditioning (ablation) | bFT · NYU v2 depth upsampling x16 | RMSE (cm): 13.36 -> 9.01 | Improved |
| use FT in every layer instead of only the final layer (ablation) | bFT · NYU v2 depth upsampling x16 | RMSE (cm): 11.40 -> 9.01 | Improved |
| apply bFT between input encoder and guide encoder instead of decoder-decoder (ablation) | bFT · NYU v2 depth upsampling x16 | RMSE (cm): 10.97 -> 9.01 | Improved |
| apply bFT between input encoder and guide encoder instead of encoder-decoder (ablation) | bFT · NYU v2 depth upsampling x16 | RMSE (cm): 11.96 -> 9.01 | Improved |
Closest papers by meaning
- Bi-Level Feature Alignment for Versatile Image Translation and Manipulation (ECCV 2022 · similarity 0.86)
- TSIT: A Simple and Versatile Framework for Image-to-Image Translation (ECCV 2020 · similarity 0.85)
- Unbalanced Feature Transport for Exemplar-Based Image Translation (CVPR 2021 · similarity 0.84)
- $\textit{Bifr\"ost}$: 3D-Aware Image Compositing with Language Instructions (NeurIPS 2024 · similarity 0.84)
- Image Translation as Diffusion Visual Programmers (ICLR 2024 · similarity 0.83)
- Learning to Manipulate Artistic Images (AAAI 2024 · similarity 0.83)
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.
