Prompt Engineering Using Detailed Text for Video Retrieval
-
- Chiba Akihiro
- Faculty of Informatics, Meisei University
-
- Suzuki Kazuya
- Faculty of Informatics, Meisei University
-
- Ueki Kazuya
- Faculty of Informatics, Meisei University
Bibliographic Information
- Other Title
-
- 映像検索における詳細テキストを用いたプロンプトエンジニアリング
Description
We propose methods to improve the accuracy of video retrieval from query texts with that describe the retrieval topic in detail as auxiliary texts for retrieval. We tested four types of prompt engineering methods: combining with query text, splitting, simplifying, and paraphrase adjustment, to obtain text that is more complementary to the query text from the original detailed text. The results of retrieval experiments using pre-trained models of CLIP showed that the three methods of combining with query sentences, simplification, and paraphrase adjustment improved retrieval accuracy, and especially, the combination of simplification and paraphrase adjustment improved the accuracy the most.
Journal
-
- Reports of the Technical Conference of the Institute of Image Electronics Engineers of Japan
-
Reports of the Technical Conference of the Institute of Image Electronics Engineers of Japan 22.04 (0), 106-109, 2023
The Institute of Image Electronics Engineers of Japan
- Tweet
Keywords
Details 詳細情報について
-
- CRID
- 1390580461190070912
-
- ISSN
- 27589218
- 02853957
-
- Text Lang
- ja
-
- Data Source
-
- JaLC
-
- Abstract License Flag
- Disallowed