Abstract: Text watermark detection in video images is challenging because text watermark characteristics are different from caption and scene texts in the video images. Developing a successful model for detecting text watermark, caption, and scene texts is an open challenge. This study aims at developing a new Deep End-to-End model for Text Watermark Detection (TWD), caption and scene text in video images. To standardize non-uniform contrast, quality, and resolution, we explore the U-Net3+ model for enhancing poor quality text without affecting high-quality text. Similarly, to address the challenges of arbitrary orientation, text shapes and complex background, we explore Stacked Hourglass Encoded Fourier Contour Embedding Network (SFCENet) by feeding the output of the U-Net3+ model as input. Furthermore, the proposed work integrates enhancement and detection models as an end-to-end model for detecting multi-type text in video images. To validate the proposed model, we create our own dataset (named TW-866), which provides video images containing text watermark, caption (subtitles), as well as scene text. The proposed model is also evaluated on standard natural scene text detection datasets, namely, ICDAR 2019 MLT, CTW1500, Total-Text, and DAST1500. The results show that the proposed method outperforms the existing methods. This is the first work on text watermark detection in video images to the best of our knowledge

Abstract: In this paper we consider the problem of incremental meta-learning in which classes are presented incrementally in discrete tasks. We propose Episodic Replay Distillation (ERD), that mixes classes from the current task with exemplars from previous tasks when sampling episodes for meta-learning. To allow the training to benefit from a large as possible variety of classes, which leads to more gener-
alizable feature representations, we propose the cross-task meta loss. Furthermore, we propose episodic replay distillation that also exploits exemplars for improved knowledge distillation. Experiments on four datasets demonstrate that ERD surpasses the state-of-the-art. In particular, on the more challenging one-shot, long task sequence scenarios, we reduce the gap between Incremental Meta-Learning and
the joint-training upper bound from 3.5% / 10.1% / 13.4% / 11.7% with the current state-of-the-art to 2.6% / 2.9% / 5.0% / 0.2% with our method on Tiered-ImageNet / Mini-ImageNet / CIFAR100 / CUB, respectively.

Abstract: Recent self-supervised learning methods are able to learn high-quality image representations and are closing the gap with supervised approaches. However, these methods are unable to acquire new knowledge incrementally – they are, in fact, mostly used only as a pre-training phase over IID data. In this work we investigate self-supervised methods in continual learning regimes without any replay
mechanism. We show that naive functional regularization,also known as feature distillation, leads to lower plasticity and limits continual learning performance. Instead, we propose Projected Functional Regularization in which a separate temporal projection network ensures that the newly learned feature space preserves information of the previous one, while at the same time allowing for the learning of new features. This prevents forgetting while maintaining the plasticity of the learner. Comparison with other incremental learning approaches applied to self-supervision demonstrates that our method obtains competitive performance in
different scenarios and on multiple datasets.

Abstract: This paper summarizes the top contributions to the first semi-supervised hyperspectral object detection (SSHOD) challenge, which was organized as a part of the Perception Beyond the Visible Spectrum (PBVS) 2022 workshop at the Computer Vision and Pattern Recognition (CVPR) conference. The SSHODC challenge is a first-of-its-kind hyperspectral dataset with temporally contiguous frames collected from a university rooftop observing a 4-way vehicle intersection over a period of three days. The dataset contains a total of 2890 frames, captured at an average resolution of 1600 × 192 pixels, with 51 hyperspectral bands from 400nm to 900nm. SSHOD challenge uses 989 images as the training set, 605 images as validation set and 1296 images as the evaluation (test) set. Each set was acquired on a different day to maximize the variance in weather conditions. Labels are provided for 10% of the annotated data, hence formulating a semi-supervised learning task for the participants which is evaluated in terms of average precision over the entire set of classes, as well as individual moving object classes: namely vehicle, bus and bike. The challenge received participation registration from 38 individuals, with 8 participating in the validation phase and 3 participating in the test phase. This paper describes the dataset acquisition, with challenge formulation, proposed methods and qualitative and quantitative results.

Abstract: In this paper, we investigate the continual learning of Vision Transformers (ViT) for the challenging exemplar-free scenario, with special focus on how to efficiently distill the knowledge of its crucial self-attention mechanism (SAM). Our work takes an initial step towards a surgical investigation of SAM for designing coherent continual learning methods in ViTs. We first carry out an evaluation of established continual learning regularization techniques. We then examine the effect of regularization when applied to two key enablers of SAM: (a) the contextualized embedding layers, for their ability to capture well-scaled representations with respect to the values, and (b) the prescaled attention maps, for carrying value-independent global contextual information. We depict the perks of each distilling strategy on two image recognition benchmarks (CIFAR100 and ImageNet-32) – while (a) leads to a better overall accuracy, (b) helps enhance the rigidity by maintaining competitive performances. Furthermore, we identify the limitation imposed by the symmetric nature of regularization losses. To alleviate this, we propose an asymmetric variant and apply it to the pooled output distillation (POD) loss adapted for ViTs. Our experiments confirm that introducing asymmetry to POD boosts its plasticity while retaining stability across (a) and (b). Moreover, we acknowledge low forgetting measures for all the compared methods, indicating that ViTs might be naturally inclined continual learners. 1

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: Deep learning has provided promising results in various applications; however, algorithms tend to be overconfident in their predictions, even though they may be entirely wrong. Particularly for critical applications, the model should provide answers only when it is very sure of them. This article presents a Bayesian version of two different state-of-the-art semantic segmentation methods to perform multi-class segmentation of foods and estimate the uncertainty about the given predictions. The proposed methods were evaluated on three public pixel-annotated food datasets. As a result, we can conclude that Bayesian methods improve the performance achieved by the baseline architectures and, in addition, provide information to improve decision-making. Furthermore, based on the extracted uncertainty map, we proposed three measures to rank the images according to the degree of noisy annotations they contained. Note that the top 135 images ranked by one of these measures include more than half of the worst-labeled food images.

Abstract: This paper proposes a novel CNN architecture for the multi-thermal image super-resolution problem. In the proposed scheme, the multi-images are synthetically generated by downsampling and slightly shifting the given image; noise is also added to each of these synthesized images. The proposed architecture uses two
attention blocks paths to extract high-frequency details taking advantage of the large information extracted from multiple images of the same scene. Experimental results are provided, showing the proposed scheme has overcome the state-of-the-art approaches.