Research map / Image generation / GANs
Incorporating Reinforced Adversarial Learning in Autoregressive Image Generation
Kenan E. Ak, Ning Xu, Zhe Lin, Yilin Wang
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 reinforced adversarial learning (single reward) | VQ-VAE with PixelCNN prior · CelebA 64x64 oracle | NLL: 7.55 -> 7.07 | Improved |
| add reinforced adversarial learning (single reward) | VQ-VAE with PixelCNN prior · CelebA 64x64 oracle | FID on oracle data: 7.49 -> 5.17 | Improved |
| add RAL (single reward) | VQ-VAE with PixelCNN prior · CelebA 64x64 | FID: 5.35 -> 3.24 | Improved |
| add RAL (single reward) | VQ-VAE-2 priors · CelebA 128x128 | FID: 58.07 -> 49.24 | Improved |
| add RAL (intermediate reward) | VQ-VAE-2 priors · CelebA 256x256 | FID: 69.37 -> 64.72 | Improved |
| add RAL (intermediate reward) | VQ-VAE-2 priors · LSUN-bedroom 128x128 | FID: 47.77 -> 36.47 | Improved |
| use intermediate reward instead of single reward | RAL · LSUN-bedroom 128x128 | FID: 39.86 -> 36.47 | Improved |
| use intermediate reward instead of single reward | RAL · CelebA 128x128 | FID: 49.24 -> 51.91 | Hurt |
| use partial generation instead of full generation (ablation) | RAL single reward · CelebA 128x128 | FID: 51.91 -> 49.24 | Improved |
| apply RAL jointly on top and bottom priors instead of top only (ablation) | RAL · CelebA 128x128 | FID: 54.82 -> 49.24 | Improved |
| apply RAL jointly on top and bottom priors instead of bottom only (ablation) | RAL · CelebA 128x128 | FID: 55.02 -> 49.24 | Improved |
| use 4x4 instead of 8x8 discriminator score map (ablation) | RAL single reward · CelebA 128x128 | FID: 53.15 -> 49.24 | Improved |
Closest papers by meaning
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.88)
- Autoregressive Adversarial Post-Training for Real-Time Interactive Video Generation (NeurIPS 2025 · similarity 0.87)
- Denoising Autoregressive Representation Learning (ICML 2024 · similarity 0.87)
- Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · similarity 0.86)
- Revisiting Adversarial Training at Scale (CVPR 2024 · similarity 0.86)
- FlexVAR: Flexible Visual Autoregressive Modeling without Residual Prediction (NeurIPS 2025 · similarity 0.86)
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.
