Research map / Image generation / Autoregressive and latent generation
GloTok: Global Perspective Tokenizer for Image Reconstruction and Generation
Xuan Zhao, Zhongyu Zhang, Yuge Huang, Yuxi Mi, Guodong Mu, Shouhong Ding, Jun Wang, Rizen Guo, Shuigeng Zhou
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 histogram relation learning to dual-codebook tokenizer (ablation) | GloTok · ImageNet-1k reconstruction ablation (4096+12288 codebooks) | rFID: 1.43 -> 0.97 | Improved |
| add histogram relation learning to dual-codebook tokenizer (ablation) | GloTok · ImageNet-1k reconstruction ablation (4096+12288 codebooks) | PSNR: 22.06 -> 22.30 | Improved |
| add residual learning module on top of histogram relation (ablation) | GloTok · ImageNet-1k reconstruction ablation (4096+12288 codebooks) | rFID: 0.97 -> 0.92 | Improved |
| add residual learning module on top of histogram relation (ablation) | GloTok · ImageNet-1k reconstruction ablation (4096+12288 codebooks) | PSNR: 22.30 -> 22.44 | Improved |
| histogram weight 0.01 instead of 0.1 (ablation) | GloTok (40 epochs, no residual) · ImageNet-1k reconstruction ablation (4096+12288 codebooks) | rFID: 1.27 -> 1.21 | Improved |
| use histogram weight 0.1 (vs none) (ablation) | GloTok (40 epochs, no residual) · ImageNet-1k reconstruction ablation (4096+12288 codebooks) | PSNR: 22.06 -> 21.97 | Hurt |
| swap teacher relation source: DINOv2-Kmeans vs FQGAN codebook (ablation) | GloTok histogram relation · ImageNet-1k reconstruction | rFID: 0.98 -> 0.95 | No clear effect |
| use GloTok tokenizer instead of FQGAN with FAR-B | FAR-B · ImageNet-1k 256x256 class-conditional generation | gFID: 3.38 -> 2.95 | Improved |
| use GloTok tokenizer instead of VQ-VAE tokenizer with xAR (ablation) | xAR-B · ImageNet-1k 256x256 class-conditional generation | gFID: 2.47 -> 2.12 | Improved |
| double codebook size to 16,384x2 (GloTok(2x)) (ablation) | GloTok + xAR-B · ImageNet-1k 256x256 class-conditional generation | gFID: 2.12 -> 2.00 | Improved |
| double codebook size to 16,384x2 (GloTok(2x)) (ablation) | GloTok + FAR-B · ImageNet-1k 256x256 class-conditional generation | gFID: 3.13 -> 2.95 | Improved |
| scale generator xAR-B to xAR-H with GloTok | GloTok + xAR · ImageNet-1k 256x256 class-conditional generation | gFID: 2.12 -> 1.75 | Improved |
Closest papers by meaning
- End-to-End Vision Tokenizer Tuning (NeurIPS 2025 · similarity 0.87)
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation (NeurIPS 2025 · similarity 0.87)
- UniTok: a Unified Tokenizer for Visual Generation and Understanding (NeurIPS 2025 · similarity 0.87)
- An Image is Worth 32 Tokens for Reconstruction and Generation (NeurIPS 2024 · similarity 0.86)
- SemHiTok: A Unified Image Tokenizer via Semantic-Guided Hierarchical Codebook for Multimodal Understanding and Generation (ICLR 2026 · similarity 0.86)
- TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation (CVPR 2025 · similarity 0.86)
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.
