Research map / Image generation / Editing and style transfer
Contrastive Learning for Diverse Disentangled Foreground Generation
Yuheng Li, Yijun Li, Jingwan Lu, Eli Shechtman, Yong Jae Lee, Krishna Kumar Singh
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 a second latent code set (known and unknown) instead of single code (ablation) | ContrasFill · Face foreground generation | KFFA: 74.71 -> 83.66 | Improved |
| add a second latent code set (known and unknown) instead of single code (ablation) | ContrasFill · Car foreground generation | KFFA: 63.09 -> 82.96 | Improved |
| use bi-modulated convolution instead of concatenating codes (ablation) | ContrasFill · Face foreground generation | KFFA: 48.99 -> 83.06 | Improved |
| use bi-modulation instead of sequential u-k modulated convolution sets (ablation) | ContrasFill · Face foreground generation | KFFA: 78.19 -> 83.06 | Hurt |
| use contrastive loss instead of reprediction loss (ablation) | ContrasFill · Face foreground generation | FID: 14.41 -> 8.41 | Improved |
| use contrastive loss instead of reprediction loss (ablation) | ContrasFill · Face foreground generation | LPIPS: 0.120 -> 0.075 | Hurt |
| ContrasFill vs CoModGAN on known-factor diversity | ContrasFill · Face foreground generation | KFFA: 52.41 -> 83.66 | Improved |
| ContrasFill vs BAT-Fill image quality | ContrasFill · Face foreground generation | FID: 15.08 -> 8.36 | Improved |
| sample known code via disentangled space vs moving along discovered direction | ContrasFill · Face KFFA over 1,000 contexts | KFFA: 56.00 -> 82.03 | Improved |
| fix known code and vary unknown codes | ContrasFill · Face KFFA over 1,000 contexts | KFFA: 82.03 -> 26.15 | Hurt |
| object-shaped mask: ContrasFill vs CoModGAN | ContrasFill · Face foreground generation (object mask) | KFFA: 51.19 -> 83.39 | Improved |
| object-shaped mask: ContrasFill vs CoModGAN | ContrasFill · Bird foreground generation (object mask) | LPIPS: 0.030 -> 0.101 | Improved |
Closest papers by meaning
- Anywhere: A Multi-Agent Framework for User-Guided, Reliable, and Diverse Foreground-Conditioned Image Generation (AAAI 2025 · similarity 0.87)
- Don't Look into the Dark: Latent Codes for Pluralistic Image Inpainting (CVPR 2024 · similarity 0.87)
- Diff-CA: Separating Common and Salient Factors with Diffusion Models (NeurIPS 2026 · similarity 0.84)
- BiGR: Harnessing Binary Latent Codes for Image Generation and Improved Visual Representation Capabilities (ICLR 2025 · similarity 0.84)
- Generative Omnimatte: Learning to Decompose Video into Layers (CVPR 2025 · similarity 0.84)
- DesignEdit: Unify Spatial-Aware Image Editing via Training-free Inpainting with a Multi-Layered Latent Diffusion Framework (AAAI 2025 · 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.
