Research map / Image generation / GANs
Correspondence Distillation from NeRF-Based GAN
Yushi Lan, Chen Change Loy, Bo Dai
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 second-order cycle consistency loss (ablation) | DDF · DatasetGAN test segmentation | mean IOU: 55.3 -> 56.9 | Improved |
| increase curriculum steps from 16 to 4096 (ablation) | DDF · evaluation set | feature similarity loss: 0.455 -> 0.287 | Improved |
| add cycle consistency loss (ablation) | DDF · self-reconstruction and texture transfer | visual quality (qualitative) | Improved |
| add deformation smoothness loss (ablation) | DDF · texture transfer | visual quality (qualitative) | Improved |
| use GAN-supervised DDF instead of 2D matching baselines | DDF vs GLU-Net · facial landmark transfer | PCK@0.01: 30.4 -> 41.6 | Improved |
| use DDF instead of SIFT Flow | DDF vs SIFT Flow · facial landmark transfer | AEPE: 5.22 -> 4.47 | Improved |
| use DDF 1-shot segmenter instead of DatasetGAN 1-shot | DDF · DatasetGAN test | mean IOU: 56.9 -> 54.6 | No clear effect |
| use CoordGAN (2D GAN) instead of 3D-aware DDF | DDF vs CoordGAN · CelebA-HQ label propagation | IOU: 45.32 -> 52.25 | Hurt |
| use early generator layers (first 5) as feature supervision (ablation) | DDF · layer-wise feature similarity | qualitative heatmap | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- CoordGAN — CoordGAN: Self-Supervised Dense Correspondences Emerge From GANs (CVPR 2022) · better on 0 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CoordGAN: Self-Supervised Dense Correspondences Emerge From GANs (CVPR 2022 · similarity 0.82)
- [Re] GANSpace: Discovering Interpretable GAN Controls (NeurIPS 2022 · similarity 0.81)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · similarity 0.81)
- Mind the Gap in Distilling StyleGANs (ECCV 2022 · similarity 0.81)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.80)
- Smoothness and Stability in GANs (ICLR 2020 · similarity 0.80)
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.
