Sorry, you need to enable JavaScript to visit this website.

facebooktwittermailshare

LEARNING GEOMETRIC FEATURES WITH DUAL-STREAM CNN FOR 3D ACTION RECOGNITION

Abstract: 

Recently, regarding several beneficial properties of depth camera, numerous 3D action recognition frameworks have studied high-level features by exploiting deep learning techniques, but nevertheless they cannot seize the meaningful characteristics of static human pose and dynamic action motion of a whole action. This paper introduces a deep network configured by two parallel streams of convolutional stacks for fully learning the deep intra-frame joint associations and inter-frame joint correlations, wherein the structure of each stream is learned from Inception-v3. In experiments, besides the compatibility verification with various backbone networks, the proposed approach achieves state-of-the-art performance in battle with several deep learning-based methods on the updated NTU RGB+D 120 dataset.

up
0 users have voted:

Comments

Question: What is network parameters the proposed method, compared with existing methods (in Table of Method Comparison in the slide)?
Answer: Thank for your interesting question. In this paper, we introduce a framework of dual-stream CNN for 3D human action recognition, in which each stream can be the transfer learning from a CNN backbone. For example, in the Table, we show the accuracy with different networks, such as GoogleNet, ResNet, and DenseNet. The network parameter includes the weights and bias of kernels in the convolutional layers and of hidden nodes in the fully connected layers. Following the network architecture of CNNs in the comparison, VGG-19 is the heaviest one. We hope that our answer can satisfy your query.

Paper Details

Authors:
Cam-Hao Hua, Nguyen Anh Tu, Dong-Seong Kim
Submitted On:
14 May 2020 - 10:53am
Short Link:
Type:
Presentation Slides
Event:
Presenter's Name:
Thien Huynh-The
Paper Code:
IVMSP-P5.5
Document Year:
2020
Cite

Document Files

This is the presentation of ICASSP paper #1133 "Learning geometric features with dual-stream CNN for 3D action recognition."

(31)

Subscribe

[1] Cam-Hao Hua, Nguyen Anh Tu, Dong-Seong Kim, "LEARNING GEOMETRIC FEATURES WITH DUAL-STREAM CNN FOR 3D ACTION RECOGNITION", IEEE SigPort, 2020. [Online]. Available: http://sigport.org/5301. Accessed: Sep. 26, 2020.
@article{5301-20,
url = {http://sigport.org/5301},
author = {Cam-Hao Hua; Nguyen Anh Tu; Dong-Seong Kim },
publisher = {IEEE SigPort},
title = {LEARNING GEOMETRIC FEATURES WITH DUAL-STREAM CNN FOR 3D ACTION RECOGNITION},
year = {2020} }
TY - EJOUR
T1 - LEARNING GEOMETRIC FEATURES WITH DUAL-STREAM CNN FOR 3D ACTION RECOGNITION
AU - Cam-Hao Hua; Nguyen Anh Tu; Dong-Seong Kim
PY - 2020
PB - IEEE SigPort
UR - http://sigport.org/5301
ER -
Cam-Hao Hua, Nguyen Anh Tu, Dong-Seong Kim. (2020). LEARNING GEOMETRIC FEATURES WITH DUAL-STREAM CNN FOR 3D ACTION RECOGNITION. IEEE SigPort. http://sigport.org/5301
Cam-Hao Hua, Nguyen Anh Tu, Dong-Seong Kim, 2020. LEARNING GEOMETRIC FEATURES WITH DUAL-STREAM CNN FOR 3D ACTION RECOGNITION. Available at: http://sigport.org/5301.
Cam-Hao Hua, Nguyen Anh Tu, Dong-Seong Kim. (2020). "LEARNING GEOMETRIC FEATURES WITH DUAL-STREAM CNN FOR 3D ACTION RECOGNITION." Web.
1. Cam-Hao Hua, Nguyen Anh Tu, Dong-Seong Kim. LEARNING GEOMETRIC FEATURES WITH DUAL-STREAM CNN FOR 3D ACTION RECOGNITION [Internet]. IEEE SigPort; 2020. Available from : http://sigport.org/5301