yolo_symbol_detector

====================== #preparing the dataset

• put the symbol images in the directory ./custom_dataset
• download or clone the repository- https://github.com/tzutalin/labelImg
• use the above tool to manually create the xml files required for training which have labels of image and coordintes of bounding box.
• after an xml file has been created for each image divide the dataset into train and test.
• put the train images and the corresponding xml files in the directory ./custom_dataset/train
• put the test images and the corresponding xml files in the directory ./custom_dataset/test
• run python file "tools/XML_to_YOLOv3.py" to generate the text files from the above xml files in the directory "model_data".
• the above step would generate three text files "symbols_names.txt", "symbols_test.txt", "symbols_train.txt" in the directory ./model_data which would be in the format required for the yolo training.

====================== #yolo training

• For training, the initial weights can be downloaded from the link: https://pjreddie.com/media/files/yolov3.weights
• include the yolov3.weights file in the directory "./model_data" which have the initial pretrained weights and the training would use these to train on the current dataset further.
• set all the necessary confugurations (the paths to "symbols_names.txt", "symbols_test.txt", "symbols_train.txt") in the "yolov3/configs.py"
• run "train.py"
• after training the weights are stored in the directory "checkpoints"
• use "tensorboard --logdir=log" and then paste the address given (http://localhost:6006/) to view the tensorboard plots

====================== #detection

• place the images having symbols to be detected in the folder "symbol_images"
• run "detection_custom.py" to use the trained model to detect the symbols in the images

====================== #some requirements

numpy>=1.18.2 scipy>=1.4.1 wget>=3.2 seaborn>=0.10.0 tensorflow==2.1.0 opencv-python==4.1.2.* tqdm==4.43.0 pandas awscli urllib3