Research map / Image generation / Autoregressive and latent generation
SCALAR: Scale-wise Controllable Visual Autoregressive Learning
Ryan Xu, Dongyang Jin, Yancheng Bai, Rui Lan, Xu Duan, Lei Sun, Xiangxiang Chu
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 |
|---|---|---|---|
| use DINOv2-S encoder instead of ViT-S (ablation) | SCALAR VAR-d12 · ImageNet-256 Canny | FID: 4.34 -> 3.12 | Improved |
| use DINOv2-S encoder instead of SAM-B (ablation) | SCALAR VAR-d12 · ImageNet-256 Canny | FID: 6.13 -> 3.12 | Improved |
| use larger DINOv2-B encoder instead of DINOv2-S (ablation) | SCALAR VAR-d12 · ImageNet-256 Depth | FID: 3.83 -> 3.68 | Improved |
| scale VAR backbone from d12 to d24 (ablation) | SCALAR · ImageNet-256 Canny | FID: 2.95 -> 2.14 | Improved |
| add Unified Control Alignment loss (ablation) | SCALAR-Uni-d12 · ImageNet-256 Canny | FID: 3.73 -> 3.31 | Improved |
| add Unified Control Alignment loss (ablation) | SCALAR-Uni-d20 · ImageNet-256 Depth | RMSE: 36.88 -> 37.12 | Hurt |
| add Unified Control Alignment loss (ablation) | SCALAR-Uni-d20 · ImageNet-256 Canny | F1-Score: 31.5 -> 31.7 | No clear effect |
| unify five conditions in one model (SCALAR-Uni vs SCALAR) | SCALAR-Uni-d12 · ImageNet-256 Depth | FID: 3.83 -> 4.21 | Hurt |
| freeze all VAR backbone parameters (ablation) | SCALAR · ImageNet-256 | control performance | Hurt |
| use SCALAR-d12 instead of ControlAR LlamaGen-L | SCALAR VAR-d12 · ImageNet-256 Canny | FID: 7.69 -> 3.12 | Improved |
| SCALAR-d24 vs ControlAR LlamaGen-L control consistency | SCALAR VAR-d24 · ImageNet-256 Canny | F1-Score: 34.9 vs 33.1 | Hurt |
| SCALAR-d24 vs ControlAR LlamaGen-L control consistency | SCALAR VAR-d24 · ImageNet-256 Depth | RMSE: 31.1 vs 33.13 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- ControlAR — ControlAR: Controllable Image Generation with Autoregressive Models (ICLR 2025) · better on 2 of 4 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
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.91)
- ControlAR: Controllable Image Generation with Autoregressive Models (ICLR 2025 · similarity 0.89)
- HMAR: Efficient Hierarchical Masked Auto-Regressive Image Generation (CVPR 2025 · similarity 0.89)
- Understand Before You Generate: Self-Guided Training for Autoregressive Image Generation (NeurIPS 2025 · similarity 0.89)
- TTS-VAR: A Test-Time Scaling Framework for Visual Auto-Regressive Generation (NeurIPS 2025 · similarity 0.88)
- Fundamental Limits of Visual Autoregressive Transformers: Universal Approximation Abilities (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.
