A5009294154- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Video Analysis and Summarization
- Domain Adaptation and Few-Shot Learning
- Image Retrieval and Classification Techniques
- Topic Modeling
- Natural Language Processing Techniques
Charles University
2019-2022
Comprehensive and fair performance evaluation of information retrieval systems represents an essential task for the current age. Whereas Cranfield-based evaluations with benchmark datasets support development models, significant efforts are required also user-oriented that try to boost interactive search approach. This article presents findings from 9th Video Browser Showdown, a competition focuses on legitimate comparison designed challenging known-item tasks over large video collection....
As reported by respected evaluation campaigns focusing both on automated and interactive video search approaches, deep learning started to dominate the retrieval area. However, results are still not satisfactory for many types of tasks high recall. To report this challenging problem, we present two orthogonal task-based performance studies centered around state-of-the-art W2VV++ query representation model retrieval. First, an ablation study is presented investigate which components effective...
We present an extended version of SOMHunter, interactive retrieval tool designed for known-item and ad-hoc search tasks over image video datasets. Even though SOMHunter has achieved significant success at the most recent Video Browser Showdown 2020 competition, lifelog datasets related constitute a different challenge with specific problems to be addressed. The presented integrates functionality required deal lots highly similar images collected in typical lifelogs. Additionally, we employ...
In the last decade, Video Browser Showdown (VBS) became a comparative platform for various interactive video search tools competing in selected retrieval tasks. However, participation of new teams with an own, novel tool is prohibitively time-demanding because large number and complexity components required constructing system from scratch. To partially alleviate this difficulty, we provide open-source version lightweight known-item SOMHunter that competed successfully at VBS 2020. The...
In the research of video retrieval systems, comparative assessments during dedicated competitions provide priceless insights into performance individual systems. The scope and depth such evaluations are unfortunately hard to improve, due limitations by set-up costs, logistics, organization complexity large events. We show that this easily impairs statistical significance collected results, reproducibility competition outcomes. article, we present a methodology for remote content-based...
SOMHunter represents a modern light-weight framework for known-item search in datasets of visual data like images or videos. The combines an effective W2VV++ text-to-image approach, traditional Bayesian model maintenance relevance scores influenced by positive examples, and several types exploration exploitation displays. With this initial setting 2020, already the first prototype system turned out to be highly competitive comparison with other state-of-the-art systems at Video Browser...
Searching for memorized images in large datasets (known-item search) is a challenging task due to limited effectiveness of retrieval models as well ability users formulate suitable queries and choose an appropriate search strategy. A popular option approach the automatically detect semantic concepts rely on interactive specification keywords during session. Nonetheless, employed instances such are often set arbitrarily existing KIS systems comprehensive evaluations with reals time demanding....