In this project we built an Encoder-Decoder model to convert images of handwritten mathematical expressions to rendable LaTeX-code. The encoder consists of a 6-layered convolutional neural network (CNN) with batch normalization and max-pooling. The decoder consists of a long short-term memory (LSTM) neural network with a soft attention mechanism. For prediction, beam search was used.
The model was trained on the CROHME dataset.
The expressions above were written by the authors themselves.