Research map / Image generation / Autoregressive and latent generation
Unleashing Transformers: Parallel Token Prediction with Discrete Absorbing Diffusion for Fast High-Resolution Image Generation from Vector-Quantized Codes
Sam Bond-Taylor, Peter Hessey, Hiroshi Sasaki, Toby P. Breckon, Chris G. Willcocks
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 |
|---|---|---|---|
| replace autoregressive prior with absorbing diffusion (ablation) | VQ image prior · LSUN Churches | FID: 5.93 -> 5.58 | Improved |
| replace autoregressive prior with absorbing diffusion (ablation) | VQ image prior · FFHQ | FID: 8.15 -> 7.12 | Improved |
| replace autoregressive prior with absorbing diffusion (ablation) | VQ image prior · LSUN Churches | NLL: 6.24 -> 6.01 | Improved |
| replace autoregressive prior with absorbing diffusion (ablation) | VQ image prior · FFHQ | NLL: 6.18 -> 5.96 | Improved |
| increase sampling steps from 50 to 256 (ablation) | Unleashing Transformers · LSUN Churches | FID: 6.86 -> 5.58 | Improved |
| increase sampling steps from 50 to 256 (ablation) | Unleashing Transformers · FFHQ | FID: 9.60 -> 7.12 | Improved |
| use sampling steps 200 instead of 256 (ablation) | Unleashing Transformers · LSUN Bedroom | FID: 5.42 -> 5.32 | No clear effect |
| add DiffAug only to VQ model training (ablation) | VQ image model · LSUN Churches | reconstruction FID: 5.25 -> 5.16 | Mixed |
| add DiffAug only to VQ model training (ablation) | VQ image model · FFHQ | reconstruction FID: 3.37 -> 6.57 | Hurt |
| add adaptive weight limit lambda_max=1 only (ablation) | VQ image model · LSUN Churches | reconstruction FID: 5.25 -> 8.67 | Hurt |
| add both DiffAug and adaptive weight limiting (ablation) | VQ image model · LSUN Churches | reconstruction FID: 5.25 -> 2.70 | Improved |
| add both DiffAug and adaptive weight limiting (ablation) | VQ image model · FFHQ | reconstruction FID: 3.37 -> 3.12 | Improved |
Closest papers by meaning
- VQ4DiT: Efficient Post-Training Vector Quantization for Diffusion Transformers (AAAI 2025 · similarity 0.89)
- SparseDiT: Token Sparsification for Efficient Diffusion Transformer (NeurIPS 2025 · similarity 0.89)
- BitDance: Scaling Autoregressive Generative Models with Binary Tokens (NeurIPS 2026 · similarity 0.88)
- Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models (CVPR 2025 · similarity 0.88)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.88)
- $\text{S}^2$Q-VDiT: Accurate Quantized Video Diffusion Transformer with Salient Data and Sparse Token Distillation (NeurIPS 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.
