Exploring Temporal Information for Improved Video Understanding

Abstract

In this dissertation, I present my work towards exploring temporal information for better video understanding. Specifically, I have worked on two problems: action recognition and semantic segmentation. For action recognition, I have proposed a framework, termed hidden two-stream networks, to learn an optimal motion representation that does not require the computation of optical flow. My framework alleviates several challenges faced in video classification, such as learning motion representations, real-time inference, multi-framerate handling, generalizability to unseen actions, etc. For semantic segmentation, I have introduced a general framework that uses video prediction models to synthesize new training samples. By scaling up the training dataset, my trained models are more accurate and robust than previous models even without modifications to the network architectures or objective functions. I believe videos have much more potential to be mined, and temporal information is one of the most important cues for machines to perceive the visual world better.

PhD dissertation. Posting here for easier access and update. Chapter 4, 6 and 7 are collaborative work

References in corpus (24)

• Distributed Representations of Words and Phrases and their Compositionality
• Two-Stream Convolutional Networks for Action Recognition in Videos
• UCF101: A Dataset of 101 Human Actions Classes From Videos in The Wild
• Caffe: Convolutional Architecture for Fast Feature Embedding
• What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?
• The Kinetics Human Action Video Dataset
• Densely Connected Convolutional Networks
• Unsupervised Learning of Video Representations using LSTMs
• Action Recognition with Trajectory-Pooled Deep-Convolutional Descriptors
• ParseNet: Looking Wider to See Better
• YouTube-8M: A Large-Scale Video Classification Benchmark
• DeMoN: Depth and Motion Network for Learning Monocular Stereo
• Towards Good Practices for Very Deep Two-Stream ConvNets
• Multi-Task Zero-Shot Action Recognition with Prioritised Data Augmentation
• The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation
• Pixel-wise Attentional Gating for Parsimonious Pixel Labeling
• Guided Optical Flow Learning
• Occlusion Aware Unsupervised Learning of Optical Flow
• Hybrid Learning of Optical Flow and Next Frame Prediction to Boost Optical Flow in the Wild
• UC Merced Submission to the ActivityNet Challenge 2016
• DAP3D-Net: Where, What and How Actions Occur in Videos?
• Improving Human Activity Recognition Through Ranking and Re-ranking
• Structured Depth Prediction in Challenging Monocular Video Sequences
• Evolution-Preserving Dense Trajectory Descriptors