Research map / Image generation / Editing and style transfer
Toward Multimodal Image-to-Image Translation
Read the paperCheck 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 |
|---|---|---|---|
| combine the cVAE-GAN and cLR-GAN objectives to constrain both latent-to-output and output-to-latent mappings | conditional image-to-image translation · Google maps → satellites | perceptual realism | Helped |
| combine the cVAE-GAN and cLR-GAN objectives | conditional image-to-image translation · Google maps → satellites | diversity (LPIPS) | Helped |
| add noise to pix2pix without an explicit latent-code constraint | pix2pix · Google maps → satellites | diversity (LPIPS) | No clear effect |
| use a deterministic encoder without KL-divergence regularization | cAE-GAN · Google maps → satellites | perceptual realism | Hurt |
| add KL-divergence regularization to the latent space | cAE-GAN · Google maps → satellites | perceptual realism | Helped |
| add prior-sampled GAN training to cVAE-GAN | cVAE-GAN++ · Google maps → satellites | perceptual realism | Helped |
| use cLR-GAN latent regression with randomly sampled latent vectors | cLR-GAN · Google maps → satellites | mode collapse: approximately 15% | Hurt |
| use E_CNN instead of E_ResNet (ablation) | BicycleGAN · labels → photos | reconstruction loss (L1): 0.292 ± 0.058 -> 0.326± 0.066 | Hurt |
| use E_CNN instead of E_ResNet (ablation) | BicycleGAN · Google maps → satellites | reconstruction loss (L1): 0.268 ± 0.070 -> 0.287 ± 0.067 | Hurt |
| inject the latent code at the input instead of at every generator layer (ablation) | BicycleGAN with E_ResNet · Google maps → satellites | reconstruction loss (L1): 0.268 ± 0.070 -> 0.266 ± 0.068 | No clear effect |
| inject the latent code at the input instead of at every generator layer (ablation) | BicycleGAN with E_ResNet · labels → photos | reconstruction loss (L1): 0.292 ± 0.058 -> 0.292 ±0.054 | No clear effect |
| use latent code dimensions 2, 8, or 256 (ablation) | BicycleGAN · labels → facades | diversity | Mixed |
Closest papers by meaning
- Latent Filter Scaling for Multimodal Unsupervised Image-To-Image Translation (CVPR 2019 · similarity 0.91)
- Multi-mapping Image-to-Image Translation via Learning Disentanglement (NeurIPS 2019 · similarity 0.90)
- Bifrost-1: Bridging Multimodal LLMs and Diffusion Models with Patch-level CLIP Latents (NeurIPS 2025 · similarity 0.87)
- Learning Visual Generative Priors without Text (CVPR 2025 · similarity 0.87)
- Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis (CVPR 2025 · similarity 0.87)
- Generative Multimodal Pretraining with Discrete Diffusion Timestep Tokens (CVPR 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.