Research map / Image generation / Autoregressive and latent generation
CSD-VAR: Content-Style Decomposition in Visual Autoregressive Models
Quang-Binh Nguyen, Minh Luu, Quang Nguyen, Anh Tran, Khoi Nguyen
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 scale-aware alternating optimization (ablation) | CSD-VAR · Validation set (35 concepts) | CSD-C: 0.501 -> 0.603 | Improved |
| add scale-aware alternating optimization (ablation) | CSD-VAR · Validation set (35 concepts) | CLIP-T: 0.270 -> 0.330 | Improved |
| add SVD-based style rectification (ablation) | CSD-VAR · Validation set (35 concepts) | DINO: 0.422 -> 0.517 | Improved |
| add SVD-based style rectification (ablation) | CSD-VAR · Validation set (35 concepts) | CSD-S: 0.503 -> 0.564 | Improved |
| add augmented K-V memory (ablation) | CSD-VAR · Validation set (35 concepts) | CLIP-I: 0.702 -> 0.751 | Improved |
| add augmented K-V memory (ablation) | CSD-VAR · Validation set (35 concepts) | CSD-S: 0.559 -> 0.564 | Improved |
| apply K-V memory to all 30 blocks instead of the first block (ablation) | CSD-VAR · Validation set (35 concepts) | CLIP-T: 0.330 -> 0.292 | Mixed |
| apply K-V memory to half of the blocks instead of the first block (ablation) | CSD-VAR · Validation set (35 concepts) | DINO: 0.517 -> 0.550 | Mixed |
| use top 10 SVD ranks instead of top 3 (ablation) | CSD-VAR · Validation set (35 concepts) | CLIP-T: 0.289 -> 0.330 | Improved |
| use top 20 SVD ranks instead of top 10 (ablation) | CSD-VAR · Validation set (35 concepts) | CLIP-T: 0.330 -> 0.302 | Hurt |
| use 4 tokens instead of 1 token (ablation) | CSD-VAR · Validation set (35 concepts) | CSD-C: 0.581 -> 0.603 | Improved |
| use 16 tokens instead of 4 tokens (ablation) | CSD-VAR · Validation set (35 concepts) | CSD-C: 0.603 -> 0.572 | Hurt |
Closest papers by meaning
- CSGO: Content-Style Composition in Text-to-Image Generation (NeurIPS 2025 · similarity 0.88)
- Conditional Balance: Improving Multi-Conditioning Trade-Offs in Image Generation (CVPR 2025 · similarity 0.87)
- Less is More: Masking Elements in Image Condition Features Avoids Content Leakages in Style Transfer Diffusion Models (ICLR 2025 · similarity 0.87)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.86)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 2025 · similarity 0.86)
- StyleSSP: Sampling StartPoint Enhancement for Training-free Diffusion-based Method for Style Transfer (CVPR 2025 · similarity 0.86)
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.
