ICACT20240357 Question.1
Questioner: tiennn18@viettel.com.vn 2024-02-01 오후 2:44:19 |
ICACT20240357 Answer.1
Answer by Auhor whkim712@kyonggi.ac.kr 2024-02-01 오후 2:44:19
Chrome  Click!! |
| Video is clipped. Can you describe more how you process this?
How to define a segment of video, the content ie. in advertises, reels,... act really fast in short amount of time, which opposite with CCTV for most of the time the frames shared the same context. Beside, subjects in a scence, within a same context, they changes action which changes the meaning of the frames, how can you classify these.
|
Our goal is not just to clip videos through Shot Boundary Detection, but to clip videos by synthesizing various information such as object information and special effects through STT and YOLO. Therefore, we will be able to complement short time content such as commercials and reels in various ways. |
ICACT20240357 Question.2
Questioner: ldaustero@bicol-u.edu.ph 2024-02-01 오후 2:46:06 |
ICACT20240357 Answer.2
Answer by Auhor whkim712@kyonggi.ac.kr 2024-02-01 오후 2:46:06
Chrome Click!! |
| Could you elaborate on how the proposed system addresses the cold start problem, especially in terms of the lack of an initial target? How does it ensure effective recommendations for new or less explored content? |
The cold start problem mentioned in the introduction simply refers to the problems encountered in current recommendation systems, and this paper aims to contribute to improving the accuracy of recommendation methods utilizing metadata by extending metadata. This means creating information about content by generating metadata through deep learning-based analysis of content in a situation where there is no information about new or less explored content. This does not guarantee a user-satisfactory recommendation system, but it can compensate for the functional deficiencies of recommendation systems. |
ICACT20240357 Question.3
Questioner: arifulislamro@gmail.com 2024-02-04 오후 5:50:11 |
ICACT20240357 Answer.3
Answer by Auhor whkim712@kyonggi.ac.kr 2024-02-04 오후 5:50:11
Chrome Click!! |
| How effective was the proposed AI-based video metadata extension model in addressing the limitations of traditional video search and recommendation systems, particularly in overcoming the challenges related to pre-added metadata, privacy concerns, and the cold start problem? Additionally, how did the model perform in terms of accuracy and relevance in providing search results and recommendations based on expanded metadata, and what considerations were taken into account for fine-tuning and optimizing its performance? |
This paper is a proposed paper, and there is no quantitative evaluation as we are currently working on it. We are currently in the process of creating a model and experimenting with additional research with data generated by the method we are proposing. |
ICACT20240357 Question.4
Questioner: songliujing@cnic.cn 2024-02-04 오후 5:47:20 |
ICACT20240357 Answer.4
Answer by Auhor whkim712@kyonggi.ac.kr 2024-02-04 오후 5:47:20
Chrome Click!! |
| How scalable is the proposed system, and does it generalize well to diverse datasets and real-world scenarios beyond the ones used in the experiments? |
There is no quantitative evaluation of this paper because it is a proposed paper and is currently being studied, but since it creates information by generating metadata for images that lack metadata through deep learning-based analysis, it cannot guarantee a satisfactory recommendation for the user, but it can prevent and compensate for the lack of functionality. It is also expected that these characteristics will be well generalized. |
ICACT20240357 Question.5
Questioner: jjhu@nttu.edu.tw 2024-02-05 오전 11:04:14 |
ICACT20240357 Answer.5
Answer by Auhor whkim712@kyonggi.ac.kr 2024-02-05 오전 11:04:14
Chrome Click!! |
| How is the scene segmentation technique you mentioned applied to video grouping and indexing? Can you explain it briefly? |
Group the information from the video and the information from the voice, and then use Shot Boundary Detection when segmenting the data for indexing. |
ICACT20240357 Question.6
Questioner: jjhu@nttu.edu.tw 2024-02-05 오전 11:06:34 |
ICACT20240357 Answer.6
Answer by Auhor whkim712@kyonggi.ac.kr 2024-02-05 오전 11:06:34
Chrome Click!! |
| How long does it take to analyze a 10-minute video? |
This paper is a proposed paper, and the proposed model is still in development, so we don't know exactly. However, time will vary depending on the object detection model you use. |
ICACT20240357 Question.7
Questioner: jjhu@nttu.edu.tw 2024-02-05 오전 11:08:15 |
ICACT20240357 Answer.7
Answer by Auhor whkim712@kyonggi.ac.kr 2024-02-05 오전 11:08:15
Chrome Click!! |
| Are the objects contained in the video highly relevant to video search and recommendation systems? |
Taking Elsa Gate as an example, there is a tag for children, but the actual video contained content such as gore and adult content. I think it is highly relevant because what I want to propose in this paper is to expand metadata through actual video content. |
