Research map / Vision-language and multimodal / Human and embodied modeling
ViQ: Text-Aligned Visual Quantized Representations at Any Resolution
Xumin Yu, Zuyan Liu, Zhenyu Yang, Yuhao Dong, Shengsheng Qian, Jiwen Lu, Han Hu, Yongming Rao
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 BN + l2 bottleneck before SimVQ (gradual regularization) (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 60.9 -> 66.6 | Improved |
| use FSQ instead of SimVQ after BN + l2 (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 66.6 -> 67.9 | Improved |
| use l-infinity instead of l2 regularization (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 67.9 -> 68.7 | Improved |
| inject RoPE with attention for position information (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 65.3 -> 68.7 | Improved |
| use RoPE instead of learnable position embedding (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 65.7 -> 68.7 | Improved |
| add self-distillation loss to text loss (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 61.3 -> 66.8 | Improved |
| add VAE-latent reconstruction loss (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 66.8 -> 68.7 | Improved |
| use VAE latent loss instead of MSE + LPIPS (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 67.0 -> 68.7 | Improved |
| use larger FSQ codebook of 128,000 (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 68.7 -> 68.3 | No clear effect |
| widen bottleneck from 128 to 1536 (ablation) | ViQ · ViQ ablation (Qwen2.5-3B) | average score: 69.1 -> 69.3 | No clear effect |
| use ViQ discrete codes instead of SigLIP2-g features for VLM training (0.5B LLM, forward time) | ViQ · VLM SFT efficiency | forward time speed-up: 70% (4k), 78% (16k) | Improved |
| use ViQ instead of UniTok quantized encoder | ViQ · Multimodal understanding (Qwen2.5-1.5B) | average score: 33.0 -> 57.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- UniTok — UniTok: a Unified Tokenizer for Visual Generation and Understanding (NeurIPS 2025) · better on 1 of 3 shared results
- QLIP — QLIP: A Dynamic Quadtree Vision Prior Enhances MLLM Performance Without Retraining (ICLR 2026) · 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 as a Dialect: Unifying Visual Understanding and Generation via Text-Aligned Representations (NeurIPS 2025 · similarity 0.89)
- From Pixels to Tokens: Byte-Pair Encoding on Quantized Visual Modalities (ICLR 2025 · similarity 0.89)
- Hybrid Token Compression for Vision-Language Models (CVPR 2026 · similarity 0.89)
- UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm (CVPR 2026 · similarity 0.89)
- Channel-wise Vector Quantization (NeurIPS 2026 · similarity 0.88)
- TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation (CVPR 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.
