Research map / Image generation / Autoregressive and latent generation
Towards Accurate Image Coding: Improved Autoregressive Image Generation With Dynamic Vector Quantization
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 DQ-Transformer_b for unconditional autoregressive generation | FFHQ · FFHQ unconditional generation | FID: 5.3 -> 4.91 | Helped |
| Use DQ-Transformer_l for class-conditional generation | ImageNet · ImageNet class-conditional generation | IS: 175.1 -> 178.2 | Helped |
| Replace fixed-length VQGAN coding with variable-length DQ-VAE coding at the same expected code length | ImageNet reconstruction · ImageNet reconstruction | rFID: 4.82 -> 4.08 | Helped |
| Assign granularities adaptively rather than randomly (ablation) | DQ-VAE · ImageNet reconstruction | rFID: 7.32 -> 4.08 | Helped |
| Increase the f=8 ratio from 0.075 to 0.1 in the tested three-granularity setup (ablation) | DQ-VAE · ImageNet reconstruction | rFID: 4.08 -> 4.96 | Hurt |
| Increase the f=8 ratio further to 0.125 in the tested three-granularity setup (ablation) | DQ-VAE · ImageNet reconstruction | rFID: 4.08 -> 6.39 | Hurt |
| Increase r_f=8 from 0 to 0.3 (ablation) | DQ-VAE with granularities {8, 16} · FFHQ reconstruction | rFID: 4.46 -> 3.02 | Helped |
| Increase r_f=8 from 0.3 to 0.5 (ablation) | DQ-VAE with granularities {8, 16} · FFHQ reconstruction | rFID: 3.02 -> 2.38 | Helped |
| Increase r_f=8 from 0.7 to 1.0 (ablation) | DQ-VAE with granularities {8, 16} · FFHQ reconstruction | rFID: 2.09 -> 1.8 | Helped |
| Increase r_f=8 from 0.7 to 1.0 (ablation) | DQ-VAE with granularities {8, 16} · FFHQ reconstruction | codebook usage: 56.01% -> 46.41% | Hurt |
| Use non-shared rather than shared content input layers (ablation) | DQ-Transformer · FFHQ unconditional generation | FID: 4.91 -> 5.54 | Hurt |
| Use shared rather than non-shared position input layers (ablation) | DQ-Transformer · FFHQ unconditional generation | FID: 4.91 -> 18.28 | Hurt |
Closest papers by meaning
- A Spark of Vision-Language Intelligence: 2-Dimensional Autoregressive Transformer for Efficient Finegrained Image Generation (ICLR 2025 · similarity 0.90)
- Channel-wise Vector Quantization (NeurIPS 2026 · similarity 0.89)
- Finite Scalar Quantization: VQ-VAE Made Simple (ICLR 2024 · similarity 0.88)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.88)
- HyperVQ: Enabling Hyperprior Entropy Modeling for VQ-Based Generative Image Compression (NeurIPS 2026 · similarity 0.87)
- Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy (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.