This repository implements CRNN[^http://arxiv.org/abs/1507.05717] with Jittor.
This repository is developed and tested on Ubuntu 20.04 and CentOS 7.
Synth90k is used for training and evaluation.
First, make sure you have transmission-cli available, which can be installed via apt install or yum install.
Then, run the following commands (Ctrl + C is necessary after transmission has completed downloading).
cd data
transmission-cli --download-dir . https://academictorrents.com/download/3d0b4f09080703d2a9c6be50715b46389fdb3af1.torrent
tar zxf mjsynth.tar.gz
mv mnt/ramdisk/max/90kDICT32px/ Synth90k
cd ..python src/create_lmdb.py Synth90kcd src
python setup.py build_ext --inplace
cd ..python src/train.pyFor more options, run:
python src/train.py -hTo evaluate a set of parameters, run:
python src/evaluate.py -r CHECKPOINT Synth90kFor more options, run:
python src/evaluate.py -hTo test the model on demo images, run:
python src/predict.py -r CHECKPOINT demo/*.jpgFor more options, run:
python src/predict.py -h