Research map / Image generation / Editing and style transfer
Image-To-Image Translation With Conditional Adversarial Networks
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 |
|---|---|---|---|
| Condition the discriminator on the input image rather than using an unconditional discriminator (ablation) | pix2pix on Cityscapes labels→photos · Cityscapes labels→photos | Class IOU: 0.01 -> 0.16 | Helped |
| Add the L1 reconstruction term to cGAN (ablation) | pix2pix on Cityscapes labels→photos · Cityscapes labels→photos | Class IOU: 0.16 -> 0.17 | Helped |
| Add the cGAN objective to L1 (ablation) | pix2pix on Cityscapes labels→photos · Cityscapes labels→photos | Class IOU: 0.11 -> 0.17 | Helped |
| Add U-Net skip connections to the encoder-decoder (ablation) | pix2pix generator on Cityscapes labels→photos · Cityscapes labels→photos | Class IOU: 0.05 -> 0.14 | Helped |
| Add U-Net skip connections when training with L1 (ablation) | pix2pix generator on Cityscapes labels→photos · Cityscapes labels→photos | Class IOU: 0.08 -> 0.13 | Helped |
| Use a 70×70 rather than a 1×1 discriminator receptive field (ablation) | pix2pix on Cityscapes labels→photos · Cityscapes labels→photos | Class IOU: 0.10 -> 0.17 | Helped |
| Increase discriminator receptive field from 16×16 to 70×70 (ablation) | pix2pix on Cityscapes labels→photos · Cityscapes labels→photos | Class IOU: 0.17 -> 0.17 | No clear effect |
| Increase discriminator receptive field from 70×70 to 286×286 (ablation) | pix2pix on Cityscapes labels→photos · Cityscapes labels→photos | Class IOU: 0.17 -> 0.11 | Hurt |
| Add L1+cGAN instead of L1 alone | pix2pix on Map→aerial photo · Map→aerial photo | Turkers labeled real: 0.8% ± 0.3% -> 18.9% ± 2.5% | Helped |
| Add cGAN to L1 for Photo→map | pix2pix · Photo→map | Turkers labeled real: 2.8% ± 1.0% -> 6.1% ± 1.3% | No clear effect |
| Use L1+cGAN rather than L2 regression for colorization | pix2pix · ImageNet colorization test split | Turkers labeled real: 16.3% ± 2.4% -> 22.5% ± 1.6% | No clear effect |
| Add cGAN to L1 for photo→labels (ablation) | pix2pix on Cityscapes · Cityscapes photo→labels | Per-pixel accuracy: 0.86 -> 0.83 | Hurt |
Closest papers by meaning
- TraVeLGAN: Image-To-Image Translation by Transformation Vector Learning (CVPR 2019 · similarity 0.91)
- Conditional Image-to-Image Translation (CVPR 2018 · similarity 0.90)
- High-Resolution Image Synthesis and Semantic Manipulation With Conditional GANs (CVPR 2018 · similarity 0.89)
- Multi-mapping Image-to-Image Translation via Learning Disentanglement (NeurIPS 2019 · similarity 0.89)
- Theoretical Analysis of Image-to-Image Translation with Adversarial Learning (ICML 2018 · similarity 0.89)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (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.