Video_Segmentation

Segment videos into groups of frames which represent a common human action. Sample video and corresponding frames have been provided.

Summary

The frames of the sample video are fed into a pre-trained Keras model of VGGNet to extract the features of the frames. The extracted features of the frames have been used for Spectral Clustering of the frames using the Normalized Cuts algorithm.

Prerequisites

1. Anaconda
2. Scikit-Learn
3. Keras alongwith Theano or Tensorflow(recommended)

How to use

1. Set the value of k to the desired number of clusters.
2. Pass the number of frames, format of the frames and the path, where the frames are located, in the get_features function
3. Pass the number of frames in the adjacency_matrix function
4. Run the rest of the code

How to extract the frames of the video

Use OpenCV

To-Do

1. Add Steps to extract the frames of the video
2. Implement Oversegmentation and add Conv3D model for finer segmentation.