1.Introduction:

We propose a novel video summarization framework named Attentive encoder-decoder networks for Video Summarization, in which the encoder uses a Bidirectional Long Short-Term Memory (BiLSTM) to encode the contextual information among the input video frames. As for the decoder, two attention-based LSTM networks.

2.Related work:

Our work is a supervised approach. Additionally, since our work applies attention-based LSTM network and LSTM is a special type of RNN, we will further review the existing RNN-based and attention-based video summarization approaches, respectively.

3.Implementation:

We propose a framework consisting of two components: an encoder-decoder model and a keyshot selection model. The encoder-decoder part measures the importance of each frame. The key shots selection model helps us to convert frame-level importance scores into shot-level scores and generating summary accounting to the threshold budget which we specify.

4.Experimental results:

This results in keyshot-based summary for the video. To minimize the number of generated keyshots, we rank the intervals based on the number of keyframes in intervals divided by their lengths, and finally apply knapsack algorithm to ensure that the produced keyshot-based summary is of maximum 15 percent in length of the original test video.

5.Conclusion:

Our learning model is a successful first step towards using human-created summaries for learning to select subsets for the challenging video summarization problem. We just scratched the surface of this fruit-bearing direction.

Reference:

• Maheswari Karlapudi posted on LinkedIn
• yashkolli/Video-Summarization-Using-Attention

Co-authors: Kolli Yashwant, Sushma Parate, Karlapudi Maheshwari