| Technical Name |
Dynamic Video Segmentation Network |
| Project Operator |
National Tsing Hua University |
| Project Host |
李濬屹 |
| Summary |
We present a detailed design of dynamic
video segmentation network (DVSNet) for fast and efficient
semantic video segmentation. DVSNet consists of two convolutional neural networks: a segmentation network and
a flow network. The former generates highly accurate se-
mantic segmentations, but is deeper and slower. The latter
is much faster than the former, but its output requires further processing to generate less accurate semantic segmentations. |
| Scientific Breakthrough |
The contributions of this work are as follows:
1. A frame division technique to apply different segmentation
strategies to different frame regions for maximizing
the usage of video redundancy and continuity.
2. A DN for determining whether to assign an input frame
region to the segmentation network, and adaptively adjusting
the update period of the key frames.
3. An adaptive key frame scheduling policy. |
| Industrial Applicability |
The main applications focused in this project are visual recognition
applications for smart robot. W focus on developing deep learning
techniques for video semantic segmentation. The techniques
developed in this project can be used not only for robotic industry but also for video surveillance industry. |
| Keyword |
Deep learning Computer vision Video surveillance Semantic segmentation Neural network Optical flow network Decision network DVSNet Real-time computation Robotic vision |
