Year: 2020
Type: dissertation
Abstract: 3-Dimensional Convolutional Neural Networks (3D ConvNets) have been adopted for videobased action recognition task recently. Many 3D ConvNets, such as C3D, I3D, and Res3D, have\nbeen proposed and achieved great success. The model ensemble techniques have been very successful in achieving better performance over a single model. But model ensemble could not be\nadopted in this case given that the single 3D ConvNets model as a base learner is unrealistic. It\nremains an open question about how to achieve better performance by leveraging multiple 3D\nConvNets models. To solve the problem, we present a two-stage framework to combine multiple\n3D ConvNets models at the feature level. In the first stage, we treat each pretrained 3D ConvNets\nmodel as a feature extractor to extract features from raw videos. We fuse the extracted features\nof different 3D ConvNets models to form the new video representation and then train a classifier\nbased on the new video representation in the second stage. We explore several widely-used feature\nfusion methods for deep features learned from different models, to learn more robust action representations from raw videos. We show that our framework outperforms any single 3D ConvNets\nmodel by a large margin and exhibits comparable performance to the state-of-the-art model on two\nvideo action recognition benchmarks.
Author: Cheng Cheng
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3208245486