Research map / Image generation / GANs
Watch Your Up-Convolution: CNN Based Generative Deep Neural Networks Are Failing to Reproduce Spectral Distributions
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 |
|---|---|---|---|
| Add spectral regularization to the generator loss | GANs using up-convolution · CelebA | spectral distribution fidelity | Helped |
| Add spectral regularization to the generator loss | DCGAN and LSGAN · CelebA, GAN training | FID stability | Helped |
| Use interpolation-based up+conv instead of transposed convolution | Generative networks · Image spectral distributions | high-frequency spectrum | Mixed |
| Add three 5x5 convolution layers after the final up-convolution (ablation) | Autoencoder with spectral loss · Spectral reconstruction | spectral distribution fidelity | Helped |
| Use one 3x3 convolution layer after up-sampling (ablation) | Autoencoder with spectral loss · Spectral reconstruction | spectral distribution fidelity | Hurt |
| Use three 3x3 convolution layers after up-sampling (ablation) | Autoencoder with spectral loss · Spectral reconstruction | spectral distribution fidelity | Hurt |
| Increase convolution filter size to 5x5 and use three layers (ablation) | Autoencoder with spectral loss · Spectral reconstruction | spectral distribution fidelity | Helped |
| Use one 5x5 convolution layer after each of the last two up-convolutions (ablation) | DCGAN with spectral regularization · CelebA, spectral distribution | spectral distribution fidelity | Mixed |
| Use two 5x5 convolution layers after the last up-convolution (ablation) | DCGAN with spectral regularization · CelebA, spectral distribution | spectral distribution fidelity | Mixed |
| Use one 5x5 convolution layer after every up-convolution (ablation) | DCGAN with spectral regularization · CelebA, spectral distribution | spectral distribution fidelity | Mixed |
| Use three 5x5 convolution layers after the last up-convolution (ablation) | DCGAN with spectral regularization · CelebA, spectral distribution | spectral distribution fidelity | Helped |
| Use azimuthally integrated power-spectrum features for fake detection | SVM classifier · Faces-HQ, 20 annotated samples | accuracy: 100% | Helped |
Closest papers by meaning
- On the Frequency Bias of Generative Models (NeurIPS 2021 · similarity 0.88)
- Spectral Distribution Aware Image Generation (AAAI 2021 · similarity 0.86)
- Fourier Spectrum Discrepancies in Deep Network Generated Images (NeurIPS 2020 · similarity 0.86)
- A Closer Look at Fourier Spectrum Discrepancies for CNN-Generated Images Detection (CVPR 2021 · similarity 0.85)
- Spectrum Translation for Refinement of Image Generation (STIG) Based on Contrastive Learning and Spectral Filter Profile (AAAI 2024 · similarity 0.85)
- Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · similarity 0.84)
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.