Research map / Image generation / Autoregressive and latent generation
DC-AR: Efficient Masked Autoregressive Image Generation with Deep Compression Hybrid Tokenizer
Yecheng Wu, Han Cai, Junyu Chen, Zhuoyang Zhang, Enze Xie, Jincheng Yu, Junsong Chen, Jinyi Hu, Yao Lu, Song Han
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 continuous residual tokens and diffusion head to discrete-only model (ablation) | DC-AR · MJHQ-30K | gFID: 6.71 -> 5.50 | Improved |
| add continuous residual tokens and diffusion head to discrete-only model (ablation) | DC-AR · GenEval | overall score: 0.66 -> 0.69 | Improved |
| add continuous residual tokens and diffusion head to discrete-only model (ablation) | DC-AR tokenizer · ImageNet 256 | rFID: 5.13 -> 1.60 | Improved |
| add continuous residual tokens and diffusion head to discrete-only model (ablation) | DC-AR · MJHQ-30K | throughput (images/s): 11.4 -> 10.3 | Hurt |
| add continuous warm-up stage before alternate fine-tuning (ablation) | DC-HT tokenizer · ImageNet | rFID: 1.76 -> 1.60 | Improved |
| freeze latent space in alternate fine-tuning instead of training all components (ablation) | DC-HT tokenizer · ImageNet | rFID: 1.92 -> 1.60 | Improved |
| pre-train at 256 then fine-tune at 512 instead of training from scratch at 512 (ablation) | DC-AR · MJHQ-30K | gFID: 6.64 -> 5.50 | Improved |
| pre-train at 256 then fine-tune at 512 instead of training from scratch at 512 (ablation) | DC-AR · MJHQ-30K | GPU hours: 1440 -> 760 | Improved |
| use discrete-token transformer with 12 sampling steps instead of continuous MAR needing 64 | DC-AR · MJHQ-30K | sampling steps: 64 -> 12 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Sana-0.6B — SANA: Efficient High-Resolution Text-to-Image Synthesis with Linear Diffusion Transformers (ICLR 2025) · better on 4 of 4 shared results
- Fluid — Fluid: Scaling Autoregressive Text-to-image Generative Models with Continuous Tokens (ICLR 2025) · better on 1 of 1 shared result
- PixArt-Sigma — PixArt-Sigma: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation (ECCV 2024) · 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
- Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy (NeurIPS 2025 · similarity 0.89)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.89)
- Hawk: Leveraging Spatial Context for Faster Autoregressive Text-to-Image Generation (NeurIPS 2025 · similarity 0.88)
- Distilled Decoding 1: One-step Sampling of Image Auto-regressive Models with Flow Matching (ICLR 2025 · similarity 0.88)
- Meissonic: Revitalizing Masked Generative Transformers for Efficient High-Resolution Text-to-Image Synthesis (ICLR 2025 · similarity 0.88)
- LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding (ICLR 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.
