Abstract: During recent years, there here has been a boom in terms of deep learning use for handwriting analysis and recognition. One main application for handwriting analysis is early detection and diagnosis in the health field. Unfortunately, most real case problems still suffer a scarcity of data, which makes difficult the use of deep learning-based models. To alleviate this problem, some works resort to synthetic data generation. Lately, more works are directed towards guided data synthetic generation, a generation that uses the domain and data knowledge to generate realistic data that can be useful to train deep learning models. In this work, we combine the domain knowledge about the Alzheimer’s disease for handwriting and use it for a more guided data generation. Concretely, we have explored the use of in-air movements for synthetic data generation.

Abstract: Sign Language Translation (SLT) is a challenging task due to its cross-domain nature, involving the translation of visual-gestural language to text. Many previous methods employ an intermediate representation, i.e., gloss sequences, to facilitate SLT, thus transforming it into a two-stage task of sign language recognition (SLR) followed by sign language translation (SLT). However, the scarcity of gloss-annotated sign language data, combined with the information bottleneck in the mid-level gloss representation, has hindered the further development of the SLT task. To address this challenge, we propose a novel Gloss-Free SLT based on Visual-Language Pretraining (GFSLT-VLP), which improves SLT by inheriting language-oriented prior knowledge from pre-trained models, without any gloss annotation assistance. Our approach involves two stages: (i) integrating Contrastive Language-Image Pre-training (CLIP) with masked self-supervised learning to create pre-tasks that bridge the semantic gap between visual and textual representations and restore masked sentences, and (ii) constructing an end-to-end architecture with an encoder-decoder-like structure that inherits the parameters of the pre-trained Visual Encoder and Text Decoder from the first stage. The seamless combination of these novel designs forms a robust sign language representation and significantly improves gloss-free sign language translation. In particular, we have achieved unprecedented improvements in terms of BLEU-4 score on the PHOENIX14T dataset (>+5) and the CSL-Daily dataset (>+3) compared to state-of-the-art gloss-free SLT methods. Furthermore, our approach also achieves competitive results on the PHOENIX14T dataset when compared with most of the gloss-based methods.

Abstract: Unsupervised representation learning aims at finding methods that learn representations from data without annotation-based signals. Abstaining from annotations not only leads to economic benefits but may – and to some extent already does – result in advantages regarding the representation’s structure, robustness, and generalizability to different tasks. In the long run, unsupervised methods are expected to surpass their supervised counterparts due to the reduction of human intervention and the inherently more general setup that does not bias the optimization towards an objective originating from specific annotation-based signals. While major advantages of unsupervised representation learning have been recently observed in natural language processing, supervised methods still dominate in vision domains for most tasks. In this dissertation, we contribute to the field of unsupervised (visual) representation learning from three perspectives: (i) Learning representations: We design unsupervised, backpropagation-free Convolutional Self-Organizing Neural Networks (CSNNs) that utilize self-organization- and Hebbian-based learning rules to learn convolutional kernels and masks to achieve deeper backpropagation-free models. Thereby, we observe that backpropagation-based and -free methods can suffer from an objective function mismatch between the unsupervised pretext task and the target task. This mismatch can lead to performance decreases for the target task. (ii) Evaluating representations: We build upon the widely used (non-)linear evaluation protocol to define pretext- and target-objective-independent metrics for measuring the objective function mismatch. With these metrics, we evaluate various pretext and target tasks and disclose dependencies of the objective function mismatch concerning different parts of the training and model setup. (iii) Transferring representations: We contribute CARLANE, the first 3-way sim-to-real domain adaptation benchmark for 2D lane detection. We adopt several well-known unsupervised domain adaptation methods as baselines and propose a method based on prototypical cross-domain self-supervised learning. Finally, we focus on pixel-based unsupervised domain adaptation and contribute a content-consistent unpaired image-to-image translation method that utilizes masks, global and local discriminators, and similarity sampling to mitigate content inconsistencies, as well as feature-attentive denormalization to fuse content-based statistics into the generator stream. In addition, we propose the cKVD metric to incorporate class-specific content inconsistencies into perceptual metrics for measuring translation quality.

Abstract: In recent years, several deep learning models have been proposed to accurately quantify and diagnose cardiac pathologies. These automated tools heavily rely on the accurate segmentation of cardiac structures in MRI images. However, segmentation of the right ventricle is challenging due to its highly complex shape and ill-defined borders. Hence, there is a need for new methods to handle such structure's geometrical and textural complexities, notably in the presence of pathologies such as Dilated Right Ventricle, Tricuspid Regurgitation, Arrhythmogenesis, Tetralogy of Fallot, and Inter-atrial Communication. The last MICCAI challenge on right ventricle segmentation was held in 2012 and included only 48 cases from a single clinical center. As part of the 12th Workshop on Statistical Atlases and Computational Models of the Heart (STACOM 2021), the M&Ms-2 challenge was organized to promote the interest of the research community around right ventricle segmentation in multi-disease, multi-view, and multi-center cardiac MRI. Three hundred sixty CMR cases, including short-axis and long-axis 4-chamber views, were collected from three Spanish hospitals using nine different scanners from three different vendors, and included a diverse set of right and left ventricle pathologies. The solutions provided by the participants show that nnU-Net achieved the best results overall. However, multi-view approaches were able to capture additional information, highlighting the need to integrate multiple cardiac diseases, views, scanners, and acquisition protocols to produce reliable automatic cardiac segmentation algorithms.

Abstract: Deep learning has made tremendous progress in the last decade due to the explosion of training data and computational power. Through end-to-end training on a
large dataset, image representations are more discriminative than the previously
used hand-crafted features. However, for many real-world applications, training
and testing on a single dataset is not realistic, as the test distribution may change over time. Continuous learning takes this situation into account, where the learner must adapt to a sequence of tasks, each with a different distribution. If you would naively continue training the model with a new task, the performance of the model would drop dramatically for the previously learned data. This phenomenon is known as catastrophic forgetting.
Many approaches have been proposed to address this problem, which can be divided into three main categories: regularization-based approaches, rehearsal-based
approaches, and parameter isolation-based approaches. However, most of the existing works focus on image classification tasks and many other computer vision tasks
have not been well-explored in the continual learning setting. Therefore, in this
thesis, we study continual learning for image generation, object re-identification,
and object counting.
For the image generation problem, since the model can generate images from the previously learned task, it is free to apply rehearsal without any limitation. We developed two methods based on generative replay. The first one uses the generated image for joint training together with the new data. The second one is based on
output pixel-wise alignment. We extensively evaluate these methods on several
benchmarks.
Next, we study continual learning for object Re-Identification (ReID). Although
most state-of-the-art methods of ReID and continual ReID use softmax-triplet loss,
we found that it is better to solve the ReID problem from a meta-learning perspective because continual learning of reID can benefit a lot from the generalization of metalearning. We also propose a distillation loss and found that the removal of the positive pairs before the distillation loss is critical.
Finally, we study continual learning for the counting problem. We study the mainstream method based on density maps and propose a new approach for density
map distillation. We found that fixing the counter head is crucial for the continual learning of object counting. To further improve results, we propose an adaptor to adapt the changing feature extractor for the fixed counter head. Extensive evaluation shows that this results in improved continual learning performance.

Abstract: Automated skill assessment in sports using video-based analysis holds great potential for revolutionizing coaching methodologies. This paper focuses on the problem of skill determination in golfers by leveraging deep learning models applied to a large database of video recordings of golf swings. We investigate different regression, ranking and classification based methods and compare to a simple baseline approach. The performance is evaluated using mean squared error (MSE) as well as computing the percentages of correctly ranked pairs based on the Kendall correlation. Our results demonstrate an improvement over the baseline, with a 35% lower mean squared error and 68% correctly ranked pairs. However, achieving fine-grained skill assessment remains challenging. This work contributes to the development of AI-driven coaching systems and advances the understanding of video-based skill determination in the context of golf.

Abstract: Despite achieving state-of-the-art performance in visual tracking, recent single-branch trackers tend to overlook the weak prior assumptions associated with the Vision Transformer (ViT) encoder and inference pipeline. Moreover, the effectiveness of discriminative trackers remains constrained due to the adoption of the dual-branch pipeline. To tackle the inferior effectiveness of the vanilla ViT, we propose an Adaptive ViT Model Prediction tracker (AViTMP) to bridge the gap between single-branch network and discriminative models. Specifically, in the proposed encoder AViT-Enc, we introduce an adaptor module and joint target state embedding to enrich the dense embedding paradigm based on ViT. Then, we combine AViT-Enc with a dense-fusion decoder and a discriminative target model to predict accurate location. Further, to mitigate the limitations of conventional inference practice, we present a novel inference pipeline called CycleTrack, which bolsters the tracking robustness in the presence of distractors via bidirectional cycle tracking verification. Lastly, we propose a dual-frame update inference strategy that adeptively handles significant challenges in long-term scenarios. In the experiments, we evaluate AViTMP on ten tracking benchmarks for a comprehensive assessment, including LaSOT, LaSOTExtSub, AVisT, etc. The experimental results unequivocally establish that AViTMP attains state-of-the-art performance, especially on long-time tracking and robustness.

Abstract: The power requirements of video-oculography systems can be prohibitive for high-speed operation on portable devices. Recently, low-power alternatives such as photosensors have been evaluated, providing gaze estimates at high frequency with a trade-off in accuracy and robustness. Potentially, an approach combining slow/high-fidelity and fast/low-fidelity sensors should be able to exploit their complementarity to track fast eye motion accurately and robustly. To foster research on this topic, we introduce OpenSFEDS, a near-eye gaze estimation dataset containing approximately 2M synthetic camera-photosensor image pairs sampled at 500 Hz under varied appearance and camera position. We also formulate the task of sensor fusion for gaze estimation, proposing a deep learning framework consisting in appearance-based encoding and temporal eye-state dynamics. We evaluate several single- and multi-rate fusion baselines on OpenSFEDS, achieving 8.7% error decrease when tracking fast eye movements with a multi-rate approach vs. a gaze forecasting approach operating with a low-speed sensor alone.