Deep Reinforcement Learning for Video Prediction
碩士 === 國立交通大學 === 資訊科學與工程研究所 === 108 === This paper leverages a classic prediction technique, known as parametric overlapped block motion compensation (POBMC), in a reinforcement learning framework for video prediction. Learning-based prediction methods with explicit motion models often suffer from...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | en_US |
Published: |
2019
|
Online Access: | http://ndltd.ncl.edu.tw/handle/nbyf54 |
id |
ndltd-TW-108NCTU5394004 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-108NCTU53940042019-11-26T05:16:54Z http://ndltd.ncl.edu.tw/handle/nbyf54 Deep Reinforcement Learning for Video Prediction 以深度增強式學習為基礎的視訊預測模型 Cho, Chuan-Yuan 卓泉源 碩士 國立交通大學 資訊科學與工程研究所 108 This paper leverages a classic prediction technique, known as parametric overlapped block motion compensation (POBMC), in a reinforcement learning framework for video prediction. Learning-based prediction methods with explicit motion models often suffer from having to estimate large numbers of motion parameters with artificial regularization. Inspired by the success of sparse motion-based prediction for video compression, we propose a parametric video prediction on a sparse motion field composed of few critical pixels and their motion vectors. The prediction is achieved by gradually refining the estimate of a future frame in iterative, discrete steps. Along the way, the identification of critical pixels and their motion estimation are addressed by two neural networks trained under a reinforcement learning setting. Our model achieves the state-of-the-art performance on CaltchPed, UCF101 and CIF datasets in one-step and multi-step prediction tests. It shows good generalization results and is able to learn well on small training data. Peng, Wen-Hsiao 彭文孝 2019 學位論文 ; thesis 38 en_US |
collection |
NDLTD |
language |
en_US |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立交通大學 === 資訊科學與工程研究所 === 108 === This paper leverages a classic prediction technique, known as parametric overlapped block motion compensation (POBMC), in a reinforcement learning framework for video prediction. Learning-based prediction methods with explicit motion models often suffer from having to estimate large numbers of motion parameters with artificial regularization. Inspired by the success of sparse motion-based prediction for video compression, we propose a parametric video prediction on a sparse motion field composed of few critical pixels and their motion vectors. The prediction is achieved by gradually refining the estimate of a future frame in iterative, discrete steps. Along the way, the identification of critical pixels and their motion estimation are addressed by two neural networks trained under a reinforcement learning setting. Our model achieves the state-of-the-art performance on CaltchPed, UCF101 and CIF datasets in one-step and multi-step prediction tests. It shows good generalization results and is able to learn well on small training data.
|
author2 |
Peng, Wen-Hsiao |
author_facet |
Peng, Wen-Hsiao Cho, Chuan-Yuan 卓泉源 |
author |
Cho, Chuan-Yuan 卓泉源 |
spellingShingle |
Cho, Chuan-Yuan 卓泉源 Deep Reinforcement Learning for Video Prediction |
author_sort |
Cho, Chuan-Yuan |
title |
Deep Reinforcement Learning for Video Prediction |
title_short |
Deep Reinforcement Learning for Video Prediction |
title_full |
Deep Reinforcement Learning for Video Prediction |
title_fullStr |
Deep Reinforcement Learning for Video Prediction |
title_full_unstemmed |
Deep Reinforcement Learning for Video Prediction |
title_sort |
deep reinforcement learning for video prediction |
publishDate |
2019 |
url |
http://ndltd.ncl.edu.tw/handle/nbyf54 |
work_keys_str_mv |
AT chochuanyuan deepreinforcementlearningforvideoprediction AT zhuōquányuán deepreinforcementlearningforvideoprediction AT chochuanyuan yǐshēndùzēngqiángshìxuéxíwèijīchǔdeshìxùnyùcèmóxíng AT zhuōquányuán yǐshēndùzēngqiángshìxuéxíwèijīchǔdeshìxùnyùcèmóxíng |
_version_ |
1719296651422597120 |