Research map / Image generation / Autoregressive and latent generation
Logit Refiner: Improving Visual Autoregressive Models via Intra-Scale Dependency Modeling
Meimingwei Li, Stefan Andreas Baumann, Felix Krause, Bjorn Ommer
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 autoregressive intra-scale sampling (causal refiner) to VAR | VAR-d16 · ImageNet 256x256 | FID: 3.30 -> 2.81 | Improved |
| use bidirectional parallel refiner with independent sampling instead of AR (ablation) | VAR-d16 + refiner · ImageNet 256x256 | FID: 3.30 -> 3.15 | No clear effect |
| continue training the base model for 30 more epochs (ablation) | VAR-d16 · ImageNet 256x256 | FID: 3.30 -> 3.12 | No clear effect |
| increase refiner depth beyond 2 blocks (ablation) | Logit Refiner · ImageNet 256x256 | FID: 2.81 -> 2.82 (d_r=4); 2.81 (d_r=8) | No clear effect |
| use depth-0 refiner (AR input projection and head only) (ablation) | Logit Refiner · ImageNet 256x256 | FID: 3.30 -> 3.02 | Improved |
| also train the base backbone jointly with the refiner (ablation) | Logit Refiner · ImageNet 256x256 | FID: 2.81 -> 2.72 | Improved |
| apply refiner at scale 2^2 (ablation) | Logit Refiner · ImageNet 256x256 | FID: 2.98 -> 2.81 | Improved |
| apply refiner at scale 8^2 (ablation) | Logit Refiner · ImageNet 256x256 | FID: 2.81 -> 2.81 | No clear effect |
| change intra-scale token ordering (ablation) | Logit Refiner · ImageNet 256x256 | FID: 2.81 vs 2.83-2.86 | No clear effect |
| add refiner for VAR-d24 versus larger VAR-d30 | VAR-d24 + refiner · ImageNet 256x256 | FID: 1.92 -> 1.83 | Improved |
| add refiner | VAR-d24 · ImageNet 256x256 | Inception Score: 312.9 -> 288.2 | Hurt |
| add refiner without CFG on small model | VAR-d16 · ImageNet 256x256 (no CFG) | FID: 3.44 -> 3.41 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Infinity-2B — Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis (CVPR 2025) · 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
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.91)
- FlexVAR: Flexible Visual Autoregressive Modeling without Residual Prediction (NeurIPS 2025 · similarity 0.89)
- Memory-Efficient Visual Autoregressive Modeling with Scale-Aware KV Cache Compression (NeurIPS 2025 · similarity 0.89)
- Edit the Bits, Diff the Codes: Bitwise Residual Editing for Visual Autoregressive Models (NeurIPS 2026 · similarity 0.89)
- Collaborative Decoding Makes Visual Auto-Regressive Modeling Efficient (CVPR 2025 · similarity 0.89)
- Parallelized Autoregressive Visual Generation (CVPR 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.
