Research map / Image generation / GANs
Large Scale Adversarial Representation Learning
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 |
|---|---|---|---|
| Use a stochastic encoder rather than a deterministic encoder (ablation) | BigBiGAN · ImageNet trainval | Cls.: 46.97 ± 0.35 -> 48.10 ± 0.13 | Helped |
| Use a standard normal latent prior instead of a uniform prior with deterministic encoding (ablation) | BigBiGAN · ImageNet trainval | Cls.: 45.11 ± 0.93 -> 48.10 ± 0.13 | Helped |
| Add the x unary loss term (ablation) | BigBiGAN · ImageNet trainval | FID: 42.92 ± 0.92 -> 31.19 ± 0.37 | Helped |
| Add the z unary loss term (ablation) | BigBiGAN without x unary · ImageNet trainval | FID: 42.92 ± 0.92 -> 39.48 ± 1.00 | Helped |
| Add the z unary loss term (ablation) | BigBiGAN with x unary · ImageNet trainval | IS: 23.19 ± 0.28 -> 22.66 ± 0.18 | Hurt |
| Increase generator capacity (ablation) | BigBiGAN · ImageNet trainval | Cls.: 43.59 ± 0.34 -> 48.10 ± 0.13 | Helped |
| Increase encoder input resolution from 128 to 256 (ablation) | BigBiGAN · ImageNet trainval | Cls.: 48.10 ± 0.13 -> 50.80 ± 0.30 | Helped |
| Increase generator output resolution from 128 to 256 (ablation) | BigBiGAN with 256-resolution encoder · ImageNet trainval | Cls.: 50.80 ± 0.30 -> 51.49 | Helped |
| Use a quadruple-width RevNet-50 encoder (ablation) | BigBiGAN with 256-resolution encoder · ImageNet trainval | Cls.: 49.42 ± 0.18 -> 57.15 | Helped |
| Increase the encoder learning rate by 10× (ablation) | BigBiGAN RevNet-50 ×4 · ImageNet trainval | Cls.: 57.15 -> 60.15 | Helped |
| Use BN+CReLU instead of AvePool features (ablation) | BigBiGAN RevNet-50 ×4 · ImageNet validation | Top-1 accuracy: 60.8 -> 61.3 | Helped |
| Add the encoder and joint discriminator to unconditional BigGAN | BigGAN + SL · ImageNet unconditional generation, 500K steps | IS: 20.4 (15.4 ± 7.57) -> 25.38 (25.33 ± 0.17) | Helped |
Closest papers by meaning
- Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · similarity 0.91)
- Transformation GAN for Unsupervised Image Synthesis and Representation Learning (CVPR 2020 · similarity 0.90)
- Self-Supervised GANs via Auxiliary Rotation Loss (CVPR 2019 · similarity 0.90)
- BigDatasetGAN: Synthesizing ImageNet With Pixel-Wise Annotations (CVPR 2022 · similarity 0.89)
- Scalable GANs with Transformers (ICML 2026 · similarity 0.88)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · 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.