trellixvulnteam/banknotes-classification-rcnn_a8fv

Region-based Convolutional Neural Network for banknotes detection

INSTALLING

1. REQUIRED DEPENDENCIES

• python 3
• tensorflow cpu or gpu >= 1.5.0

2. INSTALL DEPENDENCIES USING PIP3:

pip3 install tensorflow-gpu
pip3 install Cython
pip3 install pillow
pip3 install lxml
pip3 install jupyter
pip3 install matplotlib

3. INSTALL PROTOC

• DOWNLOAD : https://github.com/google/protobuf/releases/tag/v3.4.0
• SET ENVIRONMENT VARIABLE

SETTING UP

1. SET PATH

• REPLACE "YOUR_SYSTEM_PATH"

set PYTHONPATH=YOUR_SYSTEM_PATH/models;YOUR_SYSTEM_PATH/models/research;YOUR_SYSTEM_PATH/models/research/slim

set PATH=%PATH%;%PYTHONPATH%

2. PROTOC - GO TO models/research

protoc ./object_detection/protos/*.proto --python_out=.

3. BUILD & INSTALL

py -3 setup.py build
py -3 setup.py install

4. TEST IF IT WORKING OR NOT - GO TO models/research/object_detection

jupyter notebook object_detection_tutorial.ipynb

PREPARING TO TRAIN

1. COLLECT IMAGES

• LABEL THE IMAGES USING : https://github.com/tzutalin/labelImg
• PUT IMAGES AT models/research/object_detection/images
• DIVIDE THE IMAGES INTO train AND test

./images
--/test
--/train

2. GENERATE CSV - GO TO models/research/object_detection

py -3 xml_to_csv.py

7. GENERATE TFRECORD

• OPEN generate_tfrecord.py AND EDIT THE LABEL MAP

# TO-DO replace this with label map
def class_text_to_int(row_label):
    if row_label == 'one':
        return 1
    elif row_label == 'two':
        return 2
    elif row_label == 'three':
        return 3
    else:
        None

THEN,

py -3 generate_tfrecord.py --csv_input=images/train_labels.csv --image_dir=images/train --output_path=train.record

py -3 generate_tfrecord.py --csv_input=images/test_labels.csv --image_dir=images/test --output_path=test.record

8. CREATE label_map.pbtxt

item {
  id: 1
  name: 'one'
}

item {
  id: 2
  name: 'two'
}

item {
  id: 3
  name: 'three'
}

• EXAMPLE AT models/research/object_detection/data
• PUT label_map.pbtxt IN THE models/research/object_detection/training

9. CONFIGURE TRAINING

9.1 CONFIGURE MODEL

• COPY THE CONFIG FILE models/research/object_detection/samples/configs/XXXX.config
• PUT THE CONFIG IN THE models/research/object_detection/training

9.2 FINE TUNE MODEL

• DOWNLOAD : https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md
• EXTRACT AND PUT THE FINE TUNE MODEL IN THE models/research/object_detection

10. EDIT CONFIG FILE FROM (9.1)

• num_classes = NUM OF CLASSES
• REPLACE all "PATH_TO_BE_CONFIGURED"
• eval_config>num_examples = NUMBER OF TEST IMAGES
• max_detections_per_class = 1
• max_total_detections = 1

11. TRAINING

• REPLACE XXXX

py -3 train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/XXXX.config

12. EVALUATION

• REPLACE XXXX

py -3 eval.py --logtostderr --train_dir=training/ --pipeline_config_path=training/XXXX.config --checkpoint_dir=training/ --eval_dir=eval/

TENSORBOARD

tensorboard --logdir=training

tensorboard --logdir=eval

13. EXPORT INFERENCE GRAPH

• REPLACE XXXX and YYYY

py -3 export_inference_graph.py --input_type image_tensor --pipeline_config_path training/XXXX.config --trained_checkpoint_prefix training/model.ckpt-YYYY --output_directory inference_graph

ERRORS

• "ValueError: Tried to convert 't' to a tensor and failed. Error: Argument must be a dense tensor: range(0, 3) - got shape [3], but wanted []" = ref : EdjeElectronics/TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10#11

REF

• https://github.com/tensorflow/models/tree/master/research/object_detection
• https://github.com/EdjeElectronics/TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10