Research map / Image generation / Autoregressive and latent generation
Generative Refinement Network for Visual Synthesis
Jian Han, Jinlai Liu, Jiahuan Wang, BINGYUE PENG, Zehuan Yuan
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 global refinement instead of mask-based generation (ablation) | GRN · ImageNet 256 class-conditional | FID: 18.13 -> 3.63 | Improved |
| add global refinement with identical decoding hyperparameters (ablation) | GRN · ImageNet 256 class-conditional | FID: 185.62 -> 3.63 | Improved |
| use random selection map instead of confidence-based selection (ablation) | GRN · ImageNet 256 class-conditional | FID: 10.64 -> 3.63 | Improved |
| add complexity-aware entropy-guided sampling (ablation) | GRN_bit-B · ImageNet 256 class-conditional | FID: 3.56 -> 3.47 | Improved |
| increase HBQ rounds from 4 to 8 (16 channels) (ablation) | HBQ tokenizer · Video tokenizer reconstruction | rFVD: 163.6 -> 144.9 | Improved |
| increase latent channels from 16 to 64 (ablation) | HBQ (M=4) tokenizer · Video tokenizer reconstruction | PSNR: 30.40 -> 33.97 | Improved |
| train longer with tuned GAN weight (HBQ*) (ablation) | HBQ (M=4, 64 ch) · Video tokenizer reconstruction | rFVD: 50.6 -> 26.3 | Improved |
| raise GAN loss weight from 0.001 to 0.02 (ablation) | HBQ tokenizer (64 ch) · Video tokenizer reconstruction | rFVD / PSNR: rFVD 48.6 -> 28.6; PSNR 34.05 -> 33.73 | Mixed |
| use HBQ M=4 instead of M=2 when scaling from GRN-B to GRN-L (ablation) | GRN · ImageNet 256 class-conditional | FID | Mixed |
| predict bits instead of indices (GRN-L) (ablation) | GRN · ImageNet 256 class-conditional | FID: 2.64 -> 2.47 | Improved |
| predict bits instead of indices (GRN-B) (ablation) | GRN · ImageNet 256 class-conditional | FID | No clear effect |
| predict absolute bits instead of relative bits (flip) (ablation) | GRN_bit · ImageNet 256 class-conditional | structural stability | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Infinity — Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis (CVPR 2025) · better on 1 of 1 shared result
- MaskGIT — MaskGIT: Masked Generative Image Transformer (CVPR 2022) · better on 1 of 1 shared result
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
- BitDance: Scaling Autoregressive Generative Models with Binary Tokens (NeurIPS 2026 · similarity 0.88)
- Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models (CVPR 2025 · similarity 0.88)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.87)
- Efficient Generative Modeling with Residual Vector Quantization-Based Tokens (ICML 2025 · similarity 0.87)
- Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis (CVPR 2025 · similarity 0.87)
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation (NeurIPS 2025 · similarity 0.87)
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.
