|
|
Detekce střihů a vyhledávání známých scén ve videu s pomocí metod hlubokého učení
Souček, Tomáš ; Lokoč, Jakub (advisor) ; Peška, Ladislav (referee)
Video retrieval represents a challenging problem with many caveats and sub-problems. This thesis focuses on two of these sub-problems, namely shot transition detection and text-based search. In the case of shot detection, many solutions have been proposed over the last decades. Recently, deep learning-based approaches improved the accuracy of shot transition detection using 3D convolutional architectures and artificially created training data, but one hundred percent accuracy is still an unreachable ideal. In this thesis we present a deep network for shot transition detection TransNet V2 that reaches state-of- the-art performance on respected benchmarks. In the second case of text-based search, deep learning models projecting textual query and video frames into a joint space proved to be effective for text-based video retrieval. We investigate these query representation learning models in a setting of known-item search and propose improvements for the text encoding part of the model. 1
|
|
|
Searching Image Collections Using Deep Representations of Local Regions
Bátoryová, Jana ; Lokoč, Jakub (advisor) ; Fink, Jiří (referee)
In a known-item search task (KIS), the goal is to find a previously seen image in a multimedia collection. In this thesis, we discuss two different approaches based on the visual description of the image. In the first one, the user creates a collage of images (using images from an external search engine), based on which we provide the most similar results from the dataset. Our results show that preprocessing the images in the dataset by splitting them into several parts is a better way to work with the spatial information contained in the user input. We compared the approach to a baseline, which does not utilize this spatial information and an approach that alters a layer in a deep neural network. We also present an alternative approach to the KIS task, search by faces. In this approach, we work with the faces extracted from the images. We investigate face representation for the ability to sort the faces based on their similarity. Then we present a structure that allows easy exploration of the set of faces. We provide a demo, implementing all presented techniques.
|
|
|
Known-item search with relevance to SOM feedback
Veselý, Patrik ; Lokoč, Jakub (advisor) ; Vomlelová, Marta (referee)
Multimedia searching is usually realized by means of text search, where a large dataset is sorted with respect to a relevance to a given text query. However, if users search for just one scene or image, a sequential browsing of a larger result set is often necessary, without a guarantee that the object is found in a reasonable time. This work focuses on methods relying on relevance feedback for more effective searching in a large collection of one million images. Several relevance update and display selection approaches are compared using simulations of relevance feedback. Our experiments reveal that the investigated models are a benefit to modern multimedia search engines. 1
|
|
|
Evaluation of Keyword-Based Search Models for Known-Item Search
Mejzlík, František ; Lokoč, Jakub (advisor) ; Skopal, Tomáš (referee)
Video retrieval over large datasets is still a very challenging task, which is getting even more relevant with the rapidly growing volume of unannotated data available. Know-item search, as one of the video retrieval tasks, is limited primarily due to the limited ability of users to formulate a suitable query and low efectivity of search models. This thesis focuses mainly on selected search models based on image classifcation, which we will also compare with a commercial solution. We will examine how to transform the network output and what models to use. Also, the efect of iterative user query reformulation on overall search efectivity will be investigated. We will also present a simple simulated user model for the generation of artifcial queries and supporting software for data collection and model evaluation in a web interface. 1
|
|
|
Effective known-item search in an initial query result set in the VIRET tool
Škrhák, Vít ; Lokoč, Jakub (advisor) ; Čech, Přemysl (referee)
Modern methods for effective video retrieval combine several research areas, especi- ally similarity search, machine learning and data visualization. Selected approaches from these areas are integrated to complex search systems, which are tested/compared at in- ternational video search competitions. An example of such system is VIRET developed at KSI MFF UK. Although VIRET represents a state-of-the-art system, it is necessary to further analyze and develop ranking models and variants of interfaces for result set browsing. This bachelor thesis focuses on implementation and testing of a method for result set visualization in the 2D grid using self-organizing maps and (hierarchical) brow- sing. The implemented method is experimentally compared with sequential browsing in the VIRET tool. 1
|
|
|
Known-Item Search in Image Datasets Using Automatically Detected Keywords
Souček, Tomáš ; Lokoč, Jakub (advisor) ; Peška, Ladislav (referee)
Known-item search represents a scenario, where a user searches for one particular image in a given collection but does not know where it is located. The thesis focuses on the design and evaluation of a keyword retrieval model for known-item search in image collections. We use a deep neural network trained on a custom dataset to annotate the images. We design complex yet easy-to-use query interface for fast image retrieval. We use/design several types of artificial users to estimate the model's performance in an interactive setting. We also discuss our successful participation at two international competitions. 1
|
guest ::
