1. Installation
2. Download
3. Train
4. Test

1. Any version of tensorflow version > 1.0 should be ok.

DeepScores Dense Extended dataset needs to be preprocessed prior to training EAST for music symbol localization. In order to run the preprocessing script run pyhton preprocess.py

1. Resnet V1 50 provided by tensorflow slim: slim resnet v1 50

In order to train the model, you should provide the dataset path, in the dataset path, a separate gt text file should be provided for each image and run

python multigpu_train.py --gpu_list=0 --input_size=512 --batch_size_per_gpu=14 --checkpoint_path=/tmp/east_deepscores_resnet_v1_50_rbox/ \
--text_scale=512 --training_data_path=/data/ocr/deepscores_localization/ --geometry=RBOX --learning_rate=0.0001 --num_readers=24 \
--pretrained_model_path=/tmp/resnet_v1_50.ckpt

If you have more than one gpu, you can pass gpu ids to gpu_list(like --gpu_list=0,1,2,3) (Note: Resnet V1 50 pretrained model was used in the implementation of this paper).

run

python eval.py --test_data_path=/tmp/images/ --gpu_list=0 --checkpoint_path=/tmp/east_deepscores_resnet_v1_50_rbox/ \
--output_dir=/tmp/

a text file will be then written to the output path.

In order to calculate F1-Score, the output text files obtained from above script as well as gt files for test images are required. To obtain precision, recall and F1-Score run pyhton score.py