TokBench: Evaluating Your Visual Tokenizer before Visual Generation
Preprint
Abstract
In this work, we reveal the limitations of visual tokenizers and VAEs in preserving fine-grained features, and propose a benchmark to evaluate reconstruction performance for two challenging visual contents: text and face. Visual tokenizers and VAEs have significantly advanced visual generation and multimodal modeling by providing more efficient compressed or quantized image representations. However, while helping production models reduce computational burdens, the information loss from image compression fundamentally limits the upper bound of visual generation quality. To evaluate this upper bound, we focus on assessing reconstructed text and facial features since they typically: 1) exist at smaller scales, 2) contain dense and rich textures, 3) are prone to collapse, and 4) are highly sensitive to human vision. We first collect and curate a diverse set of clear text and face images from existing datasets. Unlike approaches using VLM models, we employ established OCR and face recognition models for evaluation, ensuring accuracy while maintaining an exceptionally lightweight assessment process requiring just 2GB memory and 4 minutes to complete. Using our benchmark, we analyze text and face reconstruction quality across various scales for different image tokenizers and VAEs. Our results show modern visual tokenizers still struggle to preserve fine-grained features, especially at smaller scales. We further extend this evaluation framework to video, conducting comprehensive analysis of video tokenizers. Additionally, we demonstrate that traditional metrics fail to accurately reflect reconstruction performance for faces and text, while our proposed metrics serve as an effective complement.Keywords
All Related Versions
This publication has 11 references indexed in Scilit:
- Liquid: Language Models are Scalable and Unified Multi-Modal GeneratorsInternational Journal of Computer Vision, 2026
- Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis2025
- TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation2025
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction2024
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding2024
- DSText V2: A comprehensive video text spotting dataset for dense and small textPattern Recognition, 2023
- Total-Text: toward orientation robustness in scene text detectionInternational Journal on Document Analysis and Recognition (IJDAR), 2019
- Curved scene text detection via transverse and longitudinal sequence connectionPattern Recognition, 2019
- FSIM: A Feature Similarity Index for Image Quality AssessmentIEEE Transactions on Image Processing, 2011
- Image quality assessment: from error visibility to structural similarityIEEE Transactions on Image Processing, 2004