Multi-Label-Prediction-For-videos-using-Zero_Shot_Learning

The dataset:

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

Feature extractor:

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

Dimension Reduction:

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

Feature embedding:

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

Feature plotting:

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.

Label prediction:

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