Research map / Vision-language and multimodal / Human and embodied modeling
VQRAE: Representation Quantization Autoencoders for Multimodal Understanding, Generation and Reconstruction
Sinan Du, Jiahao Guo, Bo Li, Shuhao Cui, Zhengzhuo Xu, Yifu Luo, Yongxian Wei, Kun Gai, Xinggang Wang, Kai Wu, Chun Yuan
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 |
|---|---|---|---|
| raise codebook dimension from 384 to 1536 (ablation) | VQRAE codebook · ImageNet-1K (stage 1 ablation, 20 epochs) | rFID: 7.69 -> 2.65 | Improved |
| raise codebook dimension from 384 to 1536 (ablation) | VQRAE codebook · ImageNet-1K (stage 1 ablation, 20 epochs) | codebook utilization: 64% -> 100% | Improved |
| raise codebook dimension from 1536 to 1920 (ablation) | VQRAE codebook · ImageNet-1K (stage 1 ablation, 20 epochs) | rFID: 2.65 -> 2.69 | No clear effect |
| use codebook dimension at most 256 (ablation) | VQRAE codebook · ImageNet-1K (stage 1 ablation, 20 epochs) | reconstruction: NA | Hurt |
| increase codebook size from 4096 to 16384 (ablation) | VQRAE codebook · ImageNet-1K (stage 1 ablation, 20 epochs) | rFID: 7.07 -> 2.65 | Improved |
| increase codebook size from 16384 to 32768 (ablation) | VQRAE codebook · ImageNet-1K (stage 1 ablation, 20 epochs) | rFID: 2.65 -> 2.78 | Hurt |
| add second-stage encoder finetuning with self-distillation (ablation) | VQRAE · ImageNet-1K val reconstruction (training strategy ablation) | rFID: 2.84 -> 2.71 | No clear effect |
| add self-distillation to end-to-end training (ablation) | VQRAE · MME-P (training strategy ablation) | MME-P: 608.9 -> 1435.2 | Improved |
| add two-stage training on top of self-distillation (ablation) | VQRAE · MME-P (training strategy ablation) | MME-P: 1435.2 -> 1439.1 | Improved |
| add self-distillation to end-to-end training (ablation) | VQRAE · ImageNet-1K val reconstruction | PSNR: 21.35 -> 19.68 | Hurt |
| add self-distillation to end-to-end training (ablation) | VQRAE · TextVQA (training strategy ablation) | accuracy: 7.0 -> 42.6 | Improved |
| add two-stage training on top of self-distillation (ablation) | VQRAE · TextVQA (training strategy ablation) | accuracy: 42.6 -> 44.0 | Improved |
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.
- TokenFlow — TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation (CVPR 2025) · 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
- TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation (CVPR 2025 · similarity 0.90)
- dRAE: Representation Autoencoder with Hyper-Spherical Codes (NeurIPS 2026 · similarity 0.90)
- Vision as a Dialect: Unifying Visual Understanding and Generation via Text-Aligned Representations (NeurIPS 2025 · similarity 0.89)
- Finite Scalar Quantization: VQ-VAE Made Simple (ICLR 2024 · similarity 0.89)
- UniFlow: A Unified Pixel Flow Tokenizer for Visual Understanding and Generation (ICLR 2026 · similarity 0.88)
- UniTok: a Unified Tokenizer for Visual Generation and Understanding (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.
