This repository contains all the code CS230 Deep Learning Final Project which uses a recurrent neural network (bi-directional LSTM) to classify dance genres among 16 different styles.
utils.py: contains utility function for dataset generation
frameToVideo.py: converts series of frames (images) into videos
subset.py: generates a subset of training and test data (each video 20 frames each)
hyperparameter_tuning.py: contains the neural network structure, plots the accuracy and loss, and tunes hyperparameters of the model
process_data.py: dataset generation for prediction model
prediction.py: prediction model
run_frame_prediction: test prediction model
base_line.py (not used anymore): contains neural network for classifying dance genres
Clone this repository and install dependecies if you are missing any.
All the datafiles should be constructed in the same directory as this repository. The Georgia Tech dataset can be accessed through here: https://www.cc.gatech.edu/cpl/projects/dance/. Only "Let' Dance: Original Frames" are used in this project.
-
Dataset generation: In
utils.py, set the image data directory, the number of frames you want to generate for each example, size of the image and whether to generate a full or small dataset. Run this file. -
Model construction and tuning: In
hyperparameter_tuning.py, set extracted dataset directory and the values of hyperparameters. Run the file.