Abstract: Instance-level segmentation of documents consists in assigning a class-aware and instance-aware label to each pixel of the image. It is a key step in document parsing for their understanding. In this paper, we present a unified transformer encoder-decoder architecture for en-to-end instance segmentation of complex layouts in document images. The method adapts a contrastive training with a mixed query selection for anchor initialization in the decoder. Later on, it performs a dot product between the obtained query embeddings and the pixel embedding map (coming from the encoder) for semantic reasoning. Extensive experimentation on competitive benchmarks like PubLayNet, PRIMA, Historical Japanese (HJ), and TableBank demonstrate that our model with SwinL backbone achieves better segmentation performance than the existing state-of-the-art approaches with the average precision of 93.72, 54.39, 84.65 and 98.04 respectively under one billion parameters. The code is made publicly available at: github.com/ayanban011/SwinDocSegmenter .

Abstract: This paper provides an overview of the DocILE 2023 Competition, its tasks, participant submissions, the competition results and possible future research directions. This first edition of the competition focused on two Information Extraction tasks, Key Information Localization and Extraction (KILE) and Line Item Recognition (LIR). Both of these tasks require detection of pre-defined categories of information in business documents. The second task additionally requires correctly grouping the information into tuples, capturing the structure laid out in the document. The competition used the recently published DocILE dataset and benchmark that stays open to new submissions. The diversity of the participant solutions indicates the potential of the dataset as the submissions included pure Computer Vision, pure Natural Language Processing, as well as multi-modal solutions and utilized all of the parts of the dataset, including the annotated, synthetic and unlabeled subsets.

Abstract: The emergence of transformer models like BERT, GPT-2, GPT-3, RoBERTa, T5 for natural language understanding tasks has opened the floodgates towards solving a wide array of machine learning tasks in other modalities like images, audio, music, sketches and so on. These language models are domain-agnostic and as a result could be applied to 1-D sequences of any kind. However, the key challenge lies in bridging the modality gap so that they could generate strong features beneficial for out-of-domain tasks. This work focuses on leveraging the power of such pre-trained language models and discusses the challenges in predicting challenging handwritten symbols and alphabets.

Abstract: In civilian video security monitoring, retrieving and tracking a person of interest often rely on witness testimony and their appearance description. Deployed systems rely on a large amount of annotated training data and are expected to show consistent performance in diverse areas and gen-eralize well between diverse settings w.r.t. different view-points, illumination, resolution, occlusions, and poses for indoor and outdoor scenes. However, for such generalization, the system would require a large amount of various an-notated data for training and evaluation. The WACV 2023 Pedestrian Attribute Recognition and Attributed-based Per-son Retrieval Challenge (UPAR-Challenge) aimed to spot-light the problem of domain gaps in a real-world surveil-lance context and highlight the challenges and limitations of existing methods. The UPAR dataset, composed of 40 important binary attributes over 12 attribute categories across four datasets, was extended with data captured from a low-flying UAV from the P-DESTRE dataset. To this aim, 0.6M additional annotations were manually labeled and vali-dated. Each track evaluated the robustness of the competing methods to domain shifts by training on limited data from a specific domain and evaluating using data from unseen do-mains. The challenge attracted 41 registered participants, but only one team managed to outperform the baseline on one track, emphasizing the task's difficulty. This work de-scribes the challenge design, the adopted dataset, obtained results, as well as future directions on the topic.

Abstract: Text and signs around roads provide crucial information for drivers, vital for safe navigation and situational awareness. Scene text recognition in motion is a challenging problem, while textual cues typically appear for a short time span, and early detection at a distance is necessary. Systems that exploit such information to assist the driver should not only extract and incorporate visual and textual cues from the video stream but also reason over time. To address this issue, we introduce RoadTextVQA, a new dataset for the task of video question answering (VideoQA) in the context of driver assistance. RoadTextVQA consists of 3, 222 driving videos collected from multiple countries, annotated with 10, 500 questions, all based on text or road signs present in the driving videos. We assess the performance of state-of-the-art video question answering models on our RoadTextVQA dataset, highlighting the significant potential for improvement in this domain and the usefulness of the dataset in advancing research on in-vehicle support systems and text-aware multimodal question answering. The dataset is available at http://cvit.iiit.ac.in/research/projects/cvit-projects/roadtextvqa.

Abstract: This paper proposes a novel model to obtain thermal image-like representations to be used as an input in any thermal image compressive sensing approach (e.g., thermal image: filtering, enhancing, super-resolution). Thermal images offer interesting information about the objects in the scene, in addition to their temperature. Unfortunately, in most of the cases thermal cameras acquire low resolution/quality images. Hence, in order to improve these images, there are several state-of-the-art approaches that exploit complementary information from a low-cost channel (visible image) to increase the image quality of an expensive channel (infrared image). In these SOTA approaches visible images are fused at different levels without paying attention the images acquire information at different bands of the spectral. In this paper a novel approach is proposed to generate thermal image-like representations from a low cost visible images, by means of a contrastive cycled GAN network. Obtained representations (synthetic thermal image) can be later on used to improve the low quality thermal image of the same scene. Experimental results on different datasets are presented.

Abstract: In this paper, we introduce ASTRA, a Transformer-based model designed for the task of Action Spotting in soccer matches. ASTRA addresses several challenges inherent in the task and dataset, including the requirement for precise action localization, the presence of a long-tail data distribution, non-visibility in certain actions, and inherent label noise. To do so, ASTRA incorporates (a) a Transformer encoder-decoder architecture to achieve the desired output temporal resolution and to produce precise predictions, (b) a balanced mixup strategy to handle the long-tail distribution of the data, (c) an uncertainty-aware displacement head to capture the label variability, and (d) input audio signal to enhance detection of non-visible actions. Results demonstrate the effectiveness of ASTRA, achieving a tight Average-mAP of 66.82 on the test set. Moreover, in the SoccerNet 2023 Action Spotting challenge, we secure the 3rd position with an Average-mAP of 70.21 on the challenge set.