Research map / Image generation / Autoregressive and latent generation
CIAR: Interval-based Collaborative Decoding for Image Generation Acceleration
Keming Ye, Zhou Zhao, Fan Wu, Shengyu Zhang
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 upper-bound loss L_u to Inter-DRO (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 47.8981 -> 24.2459 | Improved |
| add anchor loss to Inter-DRO (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 68.6400 -> 24.2459 | Improved |
| add center loss L_c (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 25.2721 -> 24.2459 | Improved |
| add lower-bound DRO loss (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 25.2849 -> 24.2459 | Improved |
| use Inter-Head instead of max-softmax confidence (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 31.1009 -> 24.2459 | Improved |
| use Inter-Head instead of random selection (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 30.1872 -> 24.2459 | Improved |
| use Inter-Head instead of entropy thresholds (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 24.5828 -> 24.2459 | Improved |
| deepen device model from 1 to 5 layers (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 24.2459 -> 23.8232 | Improved |
| deepen device model from 1 to 3 layers (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | latency speedup: x2.53 -> x2.19 | Hurt |
| increase ensemble Inter-Heads from 1 to 5 (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 24.2459 -> 24.6723 | Hurt |
| increase ensemble Inter-Heads from 1 to 3 (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 24.2459 -> 24.1301 | Improved |
| raise uncertainty threshold from 0.30 to 0.35 (ablation) | CIAR (LlamaGen Stage I) · MS-COCO val | FID: 24.2459 -> 27.1990 | Hurt |
Closest papers by meaning
- Collaborative Decoding Makes Visual Auto-Regressive Modeling Efficient (CVPR 2025 · similarity 0.90)
- LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding (ICLR 2025 · similarity 0.87)
- ZipAR: Parallel Autoregressive Image Generation through Spatial Locality (ICML 2025 · similarity 0.87)
- RandAR: Decoder-only Autoregressive Visual Generation in Random Orders (CVPR 2025 · similarity 0.86)
- Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding (ICLR 2025 · similarity 0.85)
- BiGR: Harnessing Binary Latent Codes for Image Generation and Improved Visual Representation Capabilities (ICLR 2025 · similarity 0.85)
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.
