Download presentation
Presentation is loading. Please wait.
1
Textual Video Prediction Week 2
REU Student: Emily Cosgrove Graduate Student: Amir Mazaheri Professor: Dr. Shah
2
Project Overview Video Prediction
Generative Adversarial Networks (GANs) Textual information for video prediction Goal: Enhanced video prediction
3
Progress Large Scale Movie Description Challenge (LSMDC) challenge dataset 128,000 videos Each 2-20 seconds Comes with text Clipped each video into 1 second clips (about 30 frames) Now have 359,000 video clips Text coming with each We use standard training-validation-test split (used in LSMDC ) 90% data as training, 5% validation, and 5% testing
4
Annotation: trying to lighten the mood. Someone smiles sheepishly
5
1 2 3 4 5 6 LSTM LSTM LSTM LSTM LSTM LSTM CNN CNN CNN CNN CNN CNN CNN
DE-CONV DE-CONV DE-CONV LSTM LSTM LSTM LSTM LSTM LSTM CNN CNN CNN CNN CNN CNN CNN CNN CNN CNN CNN 1 2 3 4 5 6
6
Progress Two Research Papers
Generative Adversarial Text to Image Synthesis (Scott Reed, et.) Uses GANs Developed a model to generate images based on textual description Convolutional LSTM Network: A Machine Learning Approach to Precipitation Nowcasting (Xingjian Shi, et.) Precipitation Nowcasting: challenging weather forecasting problem Proposed convolutional LSTM (ConvLSTM)
7
Next Week Make model more complete by adding Adversarial loss and text
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.