Research map / Image generation / Autoregressive and latent generation
Frequency-Aware Autoregressive Modeling for Efficient High-Resolution Image Synthesis
Zhuokun Chen, Jugang Fan, Zhuowei Yu, Bohan Zhuang, Mingkui Tan
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 |
|---|---|---|---|
| apply SparseVAR (tau=0.7) to Infinity-2B | Infinity-2B · GenEval | latency (s) and overall score: latency 2.78 -> 1.36; overall 0.7260 -> 0.7209 | Improved |
| apply SparseVAR (tau=0.6) to HART-0.7B | HART-0.7B · GenEval | latency (s) and overall score: latency 1.32 -> 0.99; overall 0.5602 -> 0.5606 | Improved |
| raise exclusion threshold tau to 0.7 (ablation) | SparseVAR on HART-0.7B · GenEval | overall score: 0.5606 -> 0.5119 | Hurt |
| raise exclusion threshold tau to 0.7 (ablation) | SparseVAR on HART-0.7B · ImageReward | score: 0.8121 -> 0.4333 | Hurt |
| apply SparseVAR (tau=0.7) to Infinity-2B | Infinity-2B · HPSv2.1 | overall score: 30.36 -> 29.89; latency 2.61 -> 1.32 | Mixed |
| add anchor tokens (retain a grid of tokens, alpha=4) (ablation) | SparseVAR · GenEval (HART, Infinity) | generation quality | Improved |
| start exclusion at a smaller stage P (ablation) | SparseVAR · GenEval | quality and latency | Hurt |
| use MSE change of block 16 instead of other blocks or logits similarity (ablation) | SparseVAR · GenEval (Infinity, tau=0.6) | generation quality | Improved |
| apply SparseVAR to non-residual FlexVAR | FlexVAR · ImageNet 512x512 | FID and latency: FID 4.33 -> 4.38; latency 0.52 -> 0.43 | Mixed |
| use token merging or selection (ToMe, PiToMe, SparseViT, ZipVL) instead of SparseVAR (ablation) | HART-0.7B · GenEval | quality vs sparsity | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- Infinity-2B — Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis (CVPR 2025) · better on 3 of 6 shared results
- HART-0.7B — HART: Efficient Visual Generation with Hybrid Autoregressive Transformer (ICLR 2025) · better on 3 of 3 shared results
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
- FlexVAR: Flexible Visual Autoregressive Modeling without Residual Prediction (NeurIPS 2025 · similarity 0.91)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.91)
- HMAR: Efficient Hierarchical Masked Auto-Regressive Image Generation (CVPR 2025 · similarity 0.89)
- NFIG: Multi-Scale Autoregressive Image Generation via Frequency Ordering (NeurIPS 2025 · similarity 0.88)
- Fundamental Limits of Visual Autoregressive Transformers: Universal Approximation Abilities (ICML 2025 · similarity 0.88)
- Visual Autoregressive Modeling for Image Super-Resolution (ICML 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.
