Research map / Image generation / Autoregressive and latent generation
Efficient Scale-Invariant Generator With Column-Row Entangled Pixel Synthesis
Thuan Hoang Nguyen, Thanh Van Le, Anh Tran
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 |
|---|---|---|---|
| Use CREPS instead of CIPS for scale-consistent synthesis | FFHQ · FFHQ scale consistency (256 to 512) | PSNR: 33.33 -> 34.65 | Improved |
| Use CREPS instead of convolution-based AnyresGAN | FFHQ · FFHQ scale consistency (256 to 512) | LPIPS: 0.07 -> 0.01 | Improved |
| Use CREPS instead of StyleGANv2 | Image generation · FFHQ-512 | FID: 3.41 -> 4.43 | Hurt |
| Use CREPS instead of CIPS | Image generation · LSUN Church-256 | FID: 2.92 -> 5.50 | Hurt |
| Add bi-line decomposition with thickness d=1 and single final fusion (ablation) | CREPS-NB · FFHQ-128 ablation | FID: 5.98 -> 11.37 | Hurt |
| Increase bi-line thickness from d=1 to d=8 with single final fusion (ablation) | CREPS with single final fusion · FFHQ-128 ablation | FID: 11.37 -> 8.23 | Improved |
| Add multiple decoders for layer-wise feature composition at d=8 (ablation) | CREPS with bi-line features · FFHQ-128 ablation | FID: 8.23 -> 4.66 | Improved |
| Remove the intermediate decoders and use additive fusion (ablation) | CREPS layer-wise feature composition · FFHQ-128 ablation | FID: 4.66 -> 6.91 | Hurt |
| Increase multiple-decoder bi-line thickness from d=4 to d=8 (ablation) | CREPS layer-wise feature composition · FFHQ-128 ablation | FID: 6.46 -> 4.66 | Improved |
Closest papers by meaning
- High-Fidelity Synthesis with Disentangled Representation (ECCV 2020 · similarity 0.86)
- Any-Resolution Training for High-Resolution Image Synthesis (ECCV 2022 · similarity 0.86)
- Arbitrary-Scale Image Synthesis (CVPR 2022 · similarity 0.85)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.84)
- Rethinking the Objectives of Vector-Quantized Tokenizers for Image Synthesis (CVPR 2024 · similarity 0.83)
- What You See is What You GAN: Rendering Every Pixel for High-Fidelity Geometry in 3D GANs (CVPR 2024 · similarity 0.83)
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.
