RaniAbboud/Automatic-Soccer-Statistics

Built an end-to-end deep learning system that identifies the team holding the ball in a soccer match at every given moment, and is thus able to accurately calculate ball possession time for each team, while existing methods (currently in use in professional matches) can only estimate it.

Final Classifiers/
	- Team_Classifier.py: the classifier which identifies the team with the ball for on-game frames
	- Frame_Type_Classifier.py: the classifier which identifies if the frame is on-game or off-game
	
	link for the weight files for the final classifiers:
		https://drive.google.com/open?id=13CRgiYysVOdFMu9hmD-UprSCMfl5UYod
	
	/data preparation/
		- make_image_directory.py: code for creating image directories (splitting the data into folders) - used for the Frame_Type_Classifier
		- make_image_directory_gamon.py: code for creating image directories (splitting the data into folders) - used for the Team_Classifier
	/demo generation/
		- generate_demo.py: code for generation the demos
	/transfer/
		- transfer_model.py: the transfer model
		- create_data_dir.py: code for creating image directories for the new data for transfer learning
		- testing_transfer.py: code for testing the transfer model
auxilary code/
	- accuracy.py: code for calculating on-game accuracy and more
/finding correlation between possession and _/
	- corr.py: code for calculating the correlation between ball possession and other important statistics of the match using a specific dataset
/labelling script/
	- label_images.py: our script that helps us label the data manually
/First Approach models [not used in final classifiers]/
	/our CNN model/
		- CNN_model2.py: our CNN model
	/VGG features 4096 model/
		- model4096.py: the model that uses VGG features (4096)
		/data preparation (extracting features)/
			- prepare4096.py: code for preparing the data and extracting featrues for the 4096 model
		/testing/
			- testing4096.py: code for testing the 4096 model
	/VGG features 73728 model/
		- model_smaller2.py: the 73728 model that uses extracted VGG features
		/data preparation (extracting features)/
			- prepare data.py: code for preparing the data and extracting featrues for the 73728 model
			- split_save_data.py: splits the data into train and validation
		/testing/
			- testing.py: code for testing the 73728 model
	/fine tuning VGG model/
		- model.py: fine tuning VGG model
		- model_without_other.py: this is when we were very close to our final classifier, here we are training only with on-game frames