Adaptive Future Frame Prediction with Ensemble Network

Future frame prediction in videos is a challenging problem because include complicated movements and large appearance changes. Learning-based future approaches have been proposed kinds of literature. A common limitation the existing learning-based mismatch training data test data. In task, we can obtain ground truth by just waiting for few frames. It means update model online phase. Then, propose an adaptive framework task. The updating consists pre-trained network, continuous-updating weight estimation network. We also show that our achieves comparable performance to state-of-the-art approaches. demonstrate approach outperforms methods especially dynamically changing scenes.