Research map / Image generation / Editing and style transfer
DeepI2I: Enabling Deep Hierarchical Image-to-Image Translation by Transferring from GANs
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 |
|---|---|---|---|
| Initialize the encoder, generator, and discriminator from a pre-trained GAN instead of training from scratch (ablation) | DeepI2I · Birds (78/per class) | mFID: 301.7 -> 146.3 | Helped |
| Initialize the encoder, generator, and discriminator from a pre-trained GAN instead of training from scratch (ablation) | DeepI2I · Foods (110/per class) | mFID: 278.2 -> 130.8 | Helped |
| Initialize the encoder, generator, and discriminator from a pre-trained GAN instead of training from scratch (ablation) | DeepI2I · Animal faces (710/per class) | mFID: 80.7 -> 68.4 | Helped |
| Initialize from BigGAN rather than train from scratch (ablation) | DeepI2I · Animal faces (71/per class; 10% data) | mFID: 280.3 -> 141.7 | Helped |
| Initialize the encoder from BigGAN instead of random initialization (ablation) | DeepI2I · Animal faces (710/per class) | mFID: 167.2 -> 68.4 | Helped |
| Add the adaptor network to align encoder and generator features (ablation) | DeepI2I · Animal faces | mFID: 92.0 -> 68.4 | Helped |
| Use hierarchical connections from multiple encoder layers rather than a single adaptor (ablation) | DeepI2I · Animal faces | image translation quality | Helped |
| Add adaptor connections from deep and shallow encoder features rather than use only deep features (ablation) | DeepI2I · Animal faces | image translation quality | Helped |
| Use more downsampling layers (ablation) | DeepI2I · Many-class image-to-image translation | image translation performance | Helped |
| Remove the reconstruction loss (ablation) | DeepI2I · Image-to-image translation | input-output structure preservation | Hurt |
| Transfer from a pre-trained StyleGAN | DeepI2I · cat2dog | image quality | Mixed |
Closest papers by meaning
- TraVeLGAN: Image-To-Image Translation by Transformation Vector Learning (CVPR 2019 · similarity 0.92)
- ReMix: Towards Image-to-Image Translation With Limited Data (CVPR 2021 · similarity 0.90)
- DA-GAN: Instance-Level Image Translation by Deep Attention Generative Adversarial Networks (CVPR 2018 · similarity 0.90)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.89)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.88)
- Compressing Image-to-Image Translation GANs Using Local Density Structures on Their Learned Manifold (AAAI 2024 · similarity 0.88)
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.