Proposing an innovative video classification and captioning system with keyframe extraction and transfer learning.
-
Addressing the growing importance of video classification with the rise in internet usage and the adoption of deep learning models. Innovation in Approach:
-
Proposing an innovative method for video classification and captioning by extracting keyframes and employing a transfer learning approach.
-
Achieving a remarkable 98% accuracy on the ISRO dataset and an impressive 99.2% on the UCF101 dataset using the proposed algorithm. Comparison with Traditional Approach:
-
Evaluating and demonstrating the superiority of the proposed algorithm over the conventional CNN+RNN architecture for video classification.
Introducing a DenseNet201+GRU based encoder-decoder model for video captioning.
Outperforming the CNN+RNN approach, the proposed model achieves a METEOR score of 0.29 on the ISRO dataset and 0.299 on the MSR-VTT dataset.