Research map / Image generation / Editing and style transfer
StEP: Style-Based Encoder Pre-Training for Multi-Modal Image Synthesis
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Pre-train a style encoder and finetune it with the generator | StEP · edges2shoes | LPIPS: 0.124 -> 0.117 | Helped |
| Pre-train a style encoder and finetune it with the generator | StEP · night2day | PSNR: 13.88 -> 15.05 | Helped |
| Pre-train the style encoder on Space Needle data instead of night2day data | StEP · night2day | PSNR: 13.75 -> 14.34 | Helped |
| Remove the VAE component (ablation) | BicycleGAN · edges2handbags | LPIPS: $0.201 \pm 0.003$ -> $0.175 \pm 0.001$ | Helped |
| Remove the VAE component (ablation) | MUNIT-p · edges2handbags | PSNR: $18.96 \pm 0.30$ -> $19.34 \pm 0.07$ | Helped |
| Replace the KL loss and Gaussian prior with L2 latent regularization (ablation) | BicycleGAN · edges2handbags | PSNR: $19.02 \pm 0.10$ -> $19.21 \pm 0.06$ | Helped |
| Replace the KL loss and Gaussian prior with L2 latent regularization (ablation) | MUNIT-p · edges2handbags | PSNR: $19.34 \pm 0.07$ -> $19.24 \pm 0.09$ | No clear effect |
| Remove cyclic reconstruction (ablation) | StEP · edges2handbags | LPIPS: $0.189 \pm 0.004$ -> $0.183 \pm 0.002$ | Helped |
| Remove random-style discriminator and latent reconstruction (ablation) | StEP · edges2handbags | LPIPS: $0.183 \pm 0.002$ -> $0.176 \pm 0.001$ | Helped |
| Remove L2 latent regularization (ablation) | StEP · edges2handbags | IS: $2.42 \pm 0.03$ -> $2.46 \pm 0.03$ | Helped |
| Use pre-trained style embeddings instead of jointly learned embeddings in a more constrained loss setup (ablation) | StEP · edges2handbags | LPIPS: $0.177 \pm 0.002$ -> $0.189 \pm 0.004$ | Hurt |
| Use Ours v2 loss setup, removing cyclic reconstruction (ablation) | StEP · edges2shoes | user preference: $50\%$ -> $41.1\%$ | Hurt |
Closest papers by meaning
- Smoothing the Disentangled Latent Style Space for Unsupervised Image-to-Image Translation (CVPR 2021 · similarity 0.91)
- Encoding in Style: A StyleGAN Encoder for Image-to-Image Translation (CVPR 2021 · similarity 0.91)
- Unpaired Image-to-Image Translation via Latent Energy Transport (CVPR 2021 · similarity 0.89)
- Multi-mapping Image-to-Image Translation via Learning Disentanglement (NeurIPS 2019 · similarity 0.89)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.87)
- Towards General Modality Translation with Contrastive and Predictive Latent Diffusion Bridge (NeurIPS 2025 · 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.