← All work

Academic / deep learning · 2023-11-10

HMER: Image to LaTeX Converter

Built a sequence-to-sequence pipeline to translate images of handwritten mathematical expressions into LaTeX.

Role
ML project developer
Contribution
Built a sequence-to-sequence pipeline to translate images of handwritten mathematical expressions into LaTeX.
Evidence
View public source ↗

Problem

Converting handwritten mathematical expressions into LaTeX requires both visual recognition and sequence generation. This academic project explores that image-to-text task.

Approach

I built a sequence-to-sequence model using a CNN/ResNet image encoder and an LSTM decoder. The implementation connects image preprocessing, model training, and generated expression output.

Evaluation

The public source includes the model and evaluation workflow. Useful checks include expression-level correctness and inspecting failures on unusual notation, layout, and handwriting.

Outcome

The project demonstrates an end-to-end deep-learning workflow for a structured image-to-text problem.

Limitations

Performance depends on dataset coverage and notation. Generated LaTeX can be incorrect and should be reviewed. The project is academic work, not evidence of production OCR reliability.