Document_MultiObject_Detection_FasterRCNN

Detect handwritten Signatures and Dates on documents with FasterRCNN in Keras

suggested directory structure: '''

data_folder ("data")
- train_images
- test_images
- train_path ("train_annotation_sig_date.txt")
- test_path ("test_annotation_sig_date.txt")
- predict_path
main_path ("Model")
- config_filename (default: "model_config.pickle")
- output_weight_path(default: "./Model/model_frcnn.hdf5')
- base_weight_path (default: "./Model/vgg16_weights_tf_dim_ordering_tf_kernels.h5")
- record_path (default: "./Model/record.csv")
model package
- results_imgs

'''

Preprocessing for new data:

Parser Argument:

yaml annotation file as in the template "new_data.yaml"

Example:

(base) C:\Users\yesit\PycharmProjects\fsf-signature_detection\Keras-FasterRCNN> python preprocessing.py --config new_data.yml

Training Process:

Parser Argument:

required argument:

-- main_path -- train_path

optional argument:

-- num_rois (default:32) -- network (default: vgg) -- [Data Augumentation options] (default: False) horizontal_flips, vertical_flips, rot_90 -- num_epochs (default: 2000) we set to 40 by our training.. -- record_path (default: none) -- config_filename (default: "./Model/model_config.pickle") -- output_weight_path (default: './Model/model_frcnn.hdf5') -- base_weight_path (default: ./Model/vgg16_weights_tf_dim_ordering_tf_kernels.h5")

example:

python train_frcnn.py --main_path "C:/Users/yesit/PycharmProjects/fsf-signature_detection\Keras-FasterRCNN" --train_path "C:\Users\yesit\PycharmProjects\fsf-signature_detection\Keras-FasterRCNN\data\train_2020.txt"

"./Model/vgg16_weights_tf_dim_ordering_tf_kernels.h5"

Testing Process:

example:

python test_frcnn.py --main_path "C:/Users/yesit/PycharmProjects/fsf-signature_detection\Keras-FasterRCNN" --test_path "C:\Users\yesit\PycharmProjects\fsf-signature_detection\Keras-FasterRCNN\data\test_2020.txt" --record_path "./MOdel/record_sig_date.csv" --model_path "C:\Users\yesit\PycharmProjects\fsf-signature_detection\Keras-FasterRCNN\Model\model_frcnn_vgg_sig_date.hdf5"

Prediction Process:

example:

python predict_frcnn.py --model_path "C:\Users\yesit\PycharmProjects\fsf-signature_detection\Keras-FasterRCNN\Model\model_frcnn_vgg_sig_date.hdf5" ----predict_images "C:\Users\yesit\PycharmProjects\fsf-signature_detection\Keras-FasterRCNN\data\predict"

svombergen / document_multiobject_detection_fasterrcnn Goto Github PK