ICACT20210124 Question.1
Questioner: contact@technologyland.co.th 2021-02-09 오후 3:23:27 |
ICACT20210124 Answer.1
Answer by Auhor js.kim@etri.re.kr 2021-02-09 오후 3:23:27
Chrome  Click!! |
| How much is the cost for installing this system? |
About 40 thousands US dollars. Thanks. |
ICACT20210124 Question.10
Questioner: skdiakffn@ajou.ac.kr 2021-02-10 오후 2:35:23 |
ICACT20210124 Answer.10
Answer by Auhor js.kim@etri.re.kr 2021-02-10 오후 2:35:23
Chrome Click!! |
| In this study, did you use a camera or sensor other than multiple high-speed (HS) cameras or multiple initial motion-unit (IMU) sensors? |
Yes, I used a single camera, not multiple HS cameras or multiple IMU sensors. |
ICACT20210124 Question.11
Questioner: rusly.eic04g@nctu.edu.tw 2021-02-11 오전 12:19:54 |
ICACT20210124 Answer.11
Answer by Auhor js.kim@etri.re.kr 2021-02-11 오전 12:19:54
Chrome Click!! |
| Thank you for your interesting presentation. How does the considered structure for CNN reflect the target feature detection, i.e. sports activity? Why you haven't considered other architectures? |
In this paper, the proposed method does not need to use CNN to detect features because it adopts initial motion data as features. |
ICACT20210124 Question.2
Questioner: 17862328070@163.com 2021-02-09 오후 2:25:30 |
ICACT20210124 Answer.2
Answer by Auhor js.kim@etri.re.kr 2021-02-09 오후 2:25:30
Chrome Click!! |
| How many kinds of human activities can you recognize in your experiment? |
In this paper, the proposed method can recognize three kinds of human activities. However, it can be trained to recognize more than three kinds of human activities without limits. |
ICACT20210124 Question.3
Questioner: kobiljonikromjanov@gmail.com 2021-02-09 오후 4:45:41 |
ICACT20210124 Answer.3
Answer by Auhor js.kim@etri.re.kr 2021-02-09 오후 4:45:41
Chrome Click!! |
| Thank you for the presentation. How did your model perform compared to the other state of art model? |
As far as I know, there is no SOTA model comparable to ours. |
ICACT20210124 Question.4
Questioner: n.brian20202@gmail.com 2021-02-09 오후 6:46:38 |
ICACT20210124 Answer.4
Answer by Auhor js.kim@etri.re.kr 2021-02-09 오후 6:46:38
Chrome Click!! |
| It was written as a feature vector with a maximum of 9 dimensions. What does the feature here represent? |
The feature vector represents init 3D position, 3D linear velocity and 3D angular velocity. |
ICACT20210124 Question.5
Questioner: linxiaocer@fzu.edu.cn 2021-02-10 오전 9:18:34 |
ICACT20210124 Answer.5
Answer by Auhor js.kim@etri.re.kr 2021-02-10 오전 9:18:34
Chrome Click!! |
| Can you give us more details about the overall procedure shown in Slide 10? |
The details of the overalll procedure are described in our paper. |
ICACT20210124 Question.6
Questioner: linxiaocer@fzu.edu.cn 2021-02-10 오전 8:58:31 |
ICACT20210124 Answer.6
Answer by Auhor js.kim@etri.re.kr 2021-02-10 오전 8:58:31
Chrome Click!! |
| The author proposed to use a single high-speed camera instead of multiple high-speed cameras, which can reduce the complexity and the cost of visual sports platform. What is the disadvantages of the proposed solution? Can you obtain a higher accuracy of human activity recognition by using multiple high-speed cameras? |
If the complexity and the cost of system do not matter, the multiple camera-based system may be better than the single camera-based one in terms of accuracy. However, the complexity and the cost of system are the most important issues when we want practical and commercial applications. |
ICACT20210124 Question.7
Questioner: linxiaocer@fzu.edu.cn 2021-02-10 오전 9:01:36 |
ICACT20210124 Answer.7
Answer by Auhor js.kim@etri.re.kr 2021-02-10 오전 9:01:36
Chrome Click!! |
| Why were the learning rate and the epochs set to 0.1 and 30? Do the different values of the learning rate and the epochs affact the accuracy? |
As far as I know, the values of the learing rate and the epochs always affact the accuracy of deep learning methods. It is a critical issue to optimally determine those values in deep learning. |
ICACT20210124 Question.8
Questioner: linxiaocer@fzu.edu.cn 2021-02-10 오전 9:04:54 |
ICACT20210124 Answer.8
Answer by Auhor js.kim@etri.re.kr 2021-02-10 오전 9:04:54
Chrome Click!! |
| The YouTube video of the presentation is absent. |
I am so sorry. But the YouTube video of the presentation is not mandatory at this conference. |
ICACT20210124 Question.9
Questioner: shaojunyi@sjtu.edu.cn 2021-02-10 오전 9:14:29 |
ICACT20210124 Answer.9
Answer by Auhor js.kim@etri.re.kr 2021-02-10 오전 9:14:29
Chrome Click!! |
| You provide the recognition accuracy of three types of sports in the experimental results. Could you further give us the trend of recognition accuracy variation with the increment of the types of sports? |
For sports games, humans generally play well-defined activities according to the types of sports. Therefore,the proposed approach using deep learning will be very effective regardless of the increment of the types of sports. |
