HTR ConvText

--- license: apache-2.0 language: - en - vi pipeline_tag: image-to-text model-index: - name: HTR-ConvText results: - task: type: image-to-text name: Handwritten Text Recognition dataset: name: IAM type: iam split: test metrics: - type: cer value: 4.0 name: Test CER - type: wer value: 12.9 name: Test WER - task: type: image-to-text name: Handwritten Text Recognition dataset: name: LAM type: lam split: test metrics: - type: cer value: 2.7 name: Test CER - type: wer value: 7.0 name: Test WER - task: type: image-to-text name: Handwritten Text Recognition dataset: name: READ2016 type: read2016 split: test metrics: - type: cer value: 3.6 name: Test CER - type: wer value: 15.7 name: Test WER - task: type: image-to-text name: Handwritten Text Recognition dataset: name: HANDS-VNOnDB type: hands-vnondb split: test metrics: - type: cer value: 3.45 name: Test CER - type: wer value: 8.9 name: Test WER --- --- # HTR-ConvText: Leveraging Convolution and Textual Information for Handwritten Text Recognition <div align="center"> <img src="image/architecture.png" alt="HTR-ConvText Architecture" width="800"/> </div> <p align="center"> <a href="https://huggingface.co/DAIR-Group/HTR-ConvText"> <img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Models-blue"> </a> <a href="https://github.com/DAIR-Group/HTR-ConvText"> <img alt="GitHub" src="https://img.shields.io/badge/GitHub-Repo-181717.svg?logo=github&logoColor=white"> </a> <a href="https://github.com/DAIR-Group/HTR-ConvText/blob/main/LICENSE"> <img alt="License" src="https://img.shields.io/badge/License-Apache%202.0-green"> </a> <a href="https://arxiv.org/abs/2512.05021"> <img alt="arXiv" src="https://img.shields.io/badge/arXiv-2512.05021-b31b1b.svg"> </a> </p> ## Highlights HTR-ConvText is a novel hybrid architecture for Handwritten Text Recognition (HTR) that effectively balances local feature extraction with global contextual modeling. Designed to overcome the limitations of standard CTC-based decoding and data-hungry Transformers, HTR-ConvText delivers state-of-the-art performance with the following key features: - **Hybrid CNN-ViT Architecture**: Seamlessly integrates a ResNet backbone with MobileViT blocks (MVP) and Conditional Posit...