Abstract: Video retrieval using natural language queries has attracted increasing interest due to its relevance in real-world applications, from intelligent access in private media galleries to web-scale video search. Learning the cross-similarity of video and text in a joint embedding space is the dominant approach. To do so, a contrastive loss is usually employed because it organizes the embedding space by putting similar items close and dissimilar items far. This framework leads to competitive recall rates, as they solely focus on the rank of the groundtruth items. Yet, assessing the quality of the ranking list is of utmost importance when considering intelligent retrieval systems, since multiple items may share similar semantics, hence a high relevance. Moreover, the aforementioned framework uses a fixed margin to separate similar and dissimilar items, treating all non-groundtruth items as equally irrelevant. In this paper we propose to use a variable margin: we argue that varying the margin used during training based on how much relevant an item is to a given query, i.e. a relevance-based margin, easily improves the quality of the ranking lists measured through nDCG and mAP. We demonstrate the advantages of our technique using different models on EPIC-Kitchens-100 and YouCook2. We show that even if we carefully tuned the fixed margin, our technique (which does not have the margin as a hyper-parameter) would still achieve better performance. Finally, extensive ablation studies and qualitative analysis support the robustness of our approach. Code will be released at \urlhttps://github.com/aranciokov/RelevanceMargin-ICMR22.

Abstract: During the last decades, the performance of optical music recognition has been increasingly improving. However, and despite the 2-dimensional nature of music notation (e.g. notes have rhythm and pitch), most works treat musical scores as a sequence of symbols in one dimension, which make their recognition still a challenge. Thus, in this work we explore the use of graph neural networks for musical score recognition. First, because graphs are suited for n-dimensional representations, and second, because the combination of graphs with deep learning has shown a great performance in similar applications. Our methodology consists of: First, we will detect each isolated/atomic symbols (those that can not be decomposed in more graphical primitives) and the primitives that form a musical symbol. Then, we will build the graph taking as root node the notehead and as leaves those primitives or symbols that modify the note’s rhythm (stem, beam, flag) or pitch (flat, sharp, natural). Finally, the graph is translated into a human-readable character sequence for a final transcription and evaluation. Our method has been tested on more than five thousand measures, showing promising results.

Abstract: Automatic waste recycling has become a very important societal challenge nowadays, raising people’s awareness for a cleaner environment and a more sustainable lifestyle. With the transition to Smart Cities, and thanks to advanced ICT solutions, this problem has received a new impulse. The waste recycling focus has shifted from general waste treating facilities to an individual responsibility, where each person should become aware of selective waste separation. The surge of the mobile devices, accompanied by a significant increase in computation power, has potentiated and facilitated this individual role. An automated image-based waste classification mechanism can help with a more efficient recycling and a reduction of contamination from residuals. Despite the good results achieved with the deep learning methodologies for this task, the Achille’s heel is that they require large neural networks which need significant computational resources for training and therefore are not suitable for mobile devices. To circumvent this apparently intractable problem, we will rely on knowledge distillation in order to transfer the network’s knowledge from a larger network (called ‘teacher’) to a smaller, more compact one, (referred as ‘student’) and thus making it possible the task of image classification on a device with limited resources. For evaluation, we considered as ‘teachers’ large architectures such as InceptionResNet or DenseNet and as ‘students’, several configurations of the MobileNets. We used the publicly available TrashNet dataset to demonstrate that the distillation process does not significantly affect system’s performance (e.g. classification accuracy) of the student network.

Abstract: Human-machine, human-robot interaction, and collaboration appear in diverse fields, from homecare to Cyber-Physical Systems. Technological development is fast, whereas real-time methods for social communication analysis that can measure small changes in sentiment and personality states, including visual, acoustic and language modalities are lagging, particularly when the goal is to build robust, appearance invariant, and fair methods. We study and compare methods capable of fusing modalities while satisfying real-time and invariant appearance conditions. We compare state-of-the-art transformer architectures in sentiment estimation and introduce them in the much less explored field of personality perception. We show that the architectures perform differently on automatic sentiment and personality perception, suggesting that each task may be better captured/modeled by a particular method. Our work calls attention to the attractive properties of the linear versions of the transformer architectures. In particular, we show that the best results are achieved by fusing the different architectures{’} preprocessing methods. However, they pose extreme conditions in computation power and energy consumption for real-time computations for quadratic transformers due to their memory requirements. In turn, linear transformers pave the way for quantifying small changes in sentiment estimation and personality perception for real-time social communications for machines and robots.

Abstract: Geometric Deep Learning has recently attracted significant interest in a wide range of machine learning fields, including document analysis. The application of Graph Neural Networks (GNNs) has become crucial in various document-related tasks since they can unravel important structural patterns, fundamental in key information extraction processes. Previous works in the literature propose task-driven models and do not take into account the full power of graphs. We propose Doc2Graph, a task-agnostic document understanding framework based on a GNN model, to solve different tasks given different types of documents. We evaluated our approach on two challenging datasets for key information extraction in form understanding, invoice layout analysis and table detection.

Abstract: In this paper, we present a framework for Multilingual Scene Text Visual Question Answering that deals with new languages in a zero-shot fashion. Specifically, we consider the task of Scene Text Visual Question Answering (STVQA) in which the question can be asked in different languages and it is not necessarily aligned to the scene text language. Thus, we first introduce a natural step towards a more generalized version of STVQA: MUST-VQA. Accounting for this, we discuss two evaluation scenarios in the constrained setting, namely IID and zero-shot and we demonstrate that the models can perform on a par on a zero-shot setting. We further provide extensive experimentation and show the effectiveness of adapting multilingual language models into STVQA tasks.

Abstract: This paper presents final results of the Out-Of-Vocabulary 2022 (OOV) challenge. The OOV contest introduces an important aspect that is not commonly studied by Optical Character Recognition (OCR) models, namely, the recognition of unseen scene text instances at training time. The competition compiles a collection of public scene text datasets comprising of 326,385 images with 4,864,405 scene text instances, thus covering a wide range of data distributions. A new and independent validation and test set is formed with scene text instances that are out of vocabulary at training time. The competition was structured in two tasks, end-to-end and cropped scene text recognition respectively. A thorough analysis of results from baselines and different participants is presented. Interestingly, current state-of-the-art models show a significant performance gap under the newly studied setting. We conclude that the OOV dataset proposed in this challenge will be an essential area to be explored in order to develop scene text models that achieve more robust and generalized predictions.