Research map / Image generation / Autoregressive and latent generation
Bridging Continuous and Discrete Tokens for Autoregressive Visual Generation
Yuqing Wang, Zhijie Lin, Yao Teng, Yuanzhi Zhu, Shuhuai Ren, Jiashi Feng, Xihui Liu
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 |
|---|---|---|---|
| predict channels autoregressively instead of in parallel (ablation) | TokenBridge · ImageNet 256 generation | gFID: 15.7 -> 1.94 | Improved |
| order channels by low-frequency energy (ablation) | TokenBridge · ImageNet 256 generation | gFID: 1.94 -> 1.89 | Improved |
| increase quantization levels B from 8 to 64 (ablation) | TokenBridge tokenizer · ImageNet reconstruction | rFID: 3.69 -> 1.11 | Improved |
| widen quantization range from +-2 to +-5 (ablation) | TokenBridge tokenizer · ImageNet reconstruction | rFID: 2.26 -> 1.11 | Improved |
| increase quantization levels B from 16 to 64 (ablation) | TokenBridge · ImageNet 256 generation | gFID: 2.03 -> 1.94 | Improved |
| group channels for joint classification (1 to 4 channels per group) (ablation) | TokenBridge · ImageNet 256 generation | gFID: 2.28 -> 3.24 | Hurt |
| group 2 channels per classification instead of 1 (ablation) | TokenBridge · ImageNet 256 generation | gFID: 2.28 -> 2.45 | Hurt |
| increase AR head capacity from 3M to 94M parameters (ablation) | TokenBridge · ImageNet 256 generation | gFID: 2.88 -> 1.94 | Improved |
| use AR head instead of diffusion head for token prediction | TokenBridge vs MAR · Token prediction time | time (ms): 311.25 -> 52.42 | Improved |
| scale TokenBridge from L to H | TokenBridge · ImageNet 256 generation | FID: 1.76 -> 1.55 | Improved |
| use post-training quantized AR tokens instead of GMM continuous tokens | TokenBridge-L · ImageNet 256 generation | FID: 3.35 -> 1.76 | Improved |
| apply quantization to other autoencoders (VAVAE, DCAE) | TokenBridge tokenizer · ImageNet reconstruction | rFID: 0.28 -> 0.28; 0.77 -> 0.77 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GIVT — GIVT: Generative Infinite-Vocabulary Transformers (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
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation (NeurIPS 2025 · similarity 0.92)
- ImageFolder: Autoregressive Image Generation with Folded Tokens (ICLR 2025 · similarity 0.92)
- Parallelized Autoregressive Visual Generation (CVPR 2025 · similarity 0.91)
- When Worse is Better: Navigating the Compression Generation Trade-off In Visual Tokenization (NeurIPS 2025 · similarity 0.91)
- Improving Autoregressive Visual Generation with Cluster-Oriented Token Prediction (CVPR 2025 · similarity 0.91)
- UniTok: a Unified Tokenizer for Visual Generation and Understanding (NeurIPS 2025 · similarity 0.90)
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.
