UCF-101 Action Dataset from the link I mentioned in the ipython file. That will directly download the dataset to the storage.

3D-CNN trained on sports 1-M dataset .so I used these trained weights to extract the video feature of UCF-101 dataset.

created a simple Nural network to reduce the dimensions of the features from 4096 to 1024 Dimensional vector.

Trained the feature vectors into the model to get semantic feature vectors.

Used Tensorboard projector to plot feature vectors into the semantic space and where we can use PCA and t-sne algorithms to classify the values.

Used my own lines of codes to predict the labels by using the feature vectors and label word vectors.