First-Person Activity Recognition Based on Three-Stream Deep Features |
Ye-Ji Kim, Dong-Gyu Lee, Seong-Whan Lee(Korea university, Korea) |
In this paper, we present a novel three-stream deep feature fusion technique to recognize interaction-level human activities from a first-person viewpoint. The proposed approach distinguishes human motion and camera ego-motion to focus on human’s movement. The features of human and camera ego-motion information are extracted from the three-stream architecture. These features are fused by considering a relationship of human action and camera ego-motion. To validate the effectiveness of our approach, we perform experiments on UTKinect-FirstPerson dataset, and achieve state-of-the-art performance. |
|
Deep Video Super-Resolution Network Using Dynamic Upsampling Filters without Explicit Motion Compensation |
Younghyun Jo, Seon Joo Kim(Yonsei University, Korea) |
Video super-resolution (VSR) has become even more important recently to provide high resolution (HR)
contents for ultra high definition displays. While many deep learning based VSR methods have been proposed, most of
them rely heavily on the accuracy of motion estimation and compensation. We introduce a fundamentally different
framework for VSR in this paper. We propose a novel end-to-end deep neural network that generates dynamic
upsampling filters and a residual image, which are computed depending on the local spatio-temporal neighborhood of
each pixel to avoid explicit motion compensation. |
|
Real-time Vehicle Detection with Learning NMS |
Inhan Kim, Dajin Kim(POSTECH, Korea) |
Real-time Vehicle Detection with Learning NMS
Inhan Kim and Dajin Kim (POSTECH, Korea)
On-road vehicles appear specific position in the image. Based on this observation, we use cropped dataset for learning network and propose the Positional Weighting (PW) block to adding spatial weights to feature map. In addition, we use the learning Non Maximum Suppression (NMS) to reduce wrong results. With these methods applied, the F-Score increases to about 2% with real-time speed. |
| |