Learning Server

Sequence

[Seq1] Run main.py

• import augmentation, learningModel, makeTFlite

[Seq2] Run augmentation.py

• image augmentation from 'org_image' folder
• save augmented images to 'aug_image' folder

[Seq3] Run learningModel.py

• start training the model
• save model weights file

[Seq4] Run makeTFlite.py

• make TFLite file for mobile app(android)
• save .tflite file to tflite_result folder

Main Python Program

1. Augmentation.py: image augmentation

• With augmentation_module.py

  • (func) check_original_pixel_coordinate

    • check pixel coordinates of original image
    • if coordinates range in [0,415] : pass
    • else coordinates range not in [0,415] : fix them and resave

  • (func) load_images_from_folder

    • load original image using cv2
    • return image arrays

  • (func) pixel_to_yolo

    • convert pixel format coordinates range in [0,415] to yolo darknet format coordinates range in [0.0,1.0]
    • check converted coordinates
    • if coordinates range in [0.0,1.0] : pass
    • else coordinates range not in [0.0,1.0] : fix them
    • return class number and darknet format coordinates

  • (func) convert_original_txt_pixel_to_yolo

    • save yolo darknet format coordinates to txt file

2. LearningModel.py

• With learningModel_module.py

  • (func) file_len

    • count the number of buildings and return it

  • (func) split_train_valid

    • split dataset: ratio train(80%) and valid(20%)

  • (func) write_config_file

    • write training config file

Open Source

1. darknet

• train yolov4-tiny model with darknet.exe
• save weights file

2. tensorflow-yolov4-tflite

• convert weights file to .tflite file

Etc