Abstract: Images and videos are pervasive in our life and communication. With advances in smart and portable devices, high capacity communication networks and high definition cinema, image and video compression are more relevant than ever. Traditional block-based linear transform codecs such as JPEG, H.264/AVC or the recent H.266/VVC are carefully designed to meet not only the rate-distortion criteria, but also the practical requirements of applications.
Recently, a new paradigm based on deep neural networks (i.e., neural image/video compression) has become increasingly popular due to its ability to learn powerful nonlinear transforms and other coding tools directly from data instead of being crafted by humans, as was usual in previous coding formats. While achieving excellent rate-distortion performance, these approaches are still limited mostly to research environments due to heavy models and other practical limitations, such as being limited to function on a particular rate and due to high memory and computational cost. In this thesis, we study these practical limitations, and designing more practical neural image compression approaches.
After analyzing the differences between traditional and neural image compression, our first contribution is the modulated autoencoder (MAE), a framework that includes a mechanism to provide multiple rate-distortion options within a single model with comparable performance to independent models. In a second contribution, we propose the slimmable compressive autoencoder (SlimCAE), which in addition to variable rate, can optimize the complexity of the model and thus reduce significantly the memory and computational burden.
Modern generative models can learn custom image transformation directly from suitable datasets following encoder-decoder architectures, task known as image-to-image (I2I) translation. Building on our previous work, we study the problem of distributed I2I translation, where the latent representation is transmitted through a binary channel and decoded in a remote receiving side. We also propose a variant that can perform both translation and the usual autoencoding functionality.
Finally, we also consider neural video compression, where the autoencoder is typically augmented with temporal prediction via motion compensation. One of the main bottlenecks of that framework is the optical flow module that estimates the displacement to predict the next frame. Focusing on this module, we propose a method that improves the accuracy of the optical flow estimation and a simplified variant that reduces the computational cost.
Key words: neural image compression, neural video compression, optical flow, practical neural image compression, compressive autoencoders, image-to-image translation, deep learning.

Abstract: Artificial intelligence is innovating the fashion industry by proposing new applications and solutions to the problems encountered by researchers and engineers working in the industry. In this thesis, we address three of these problems. In the first part of the thesis, we tackle the problem of multi-label image classification which is very related to fashion attribute recognition. In the second part of the thesis, we address two problems that are specific to fashion. Firstly, we address the problem of main product detection which is the task of associating correct image parts (e.g. bounding boxes) with the fashion product being sold. Secondly, we address the problem of color naming for multicolored fashion items. The task of multi-label image classification consists in assigning various concepts such as objects or attributes to images. Usually, there are dependencies that can be learned between the concepts to capture label correlations (chair and table classes are more likely to co-exist than chair and giraffe).
If we treat the multi-label image classification problem as an orderless set prediction problem, we can exploit recurrent neural networks (RNN) to capture label correlations. However, RNNs are trained to predict ordered sequences of tokens, so if the order of the predicted sequence is different than the order of the ground truth sequence, there will be penalization although the predictions are correct. Therefore, in the first part of the thesis, we propose an orderless loss function which will order the labels in the ground truth sequence dynamically in a way that the minimum loss is achieved. This results in a significant improvement of RNN models on multi-label image classification over the previous methods.
However, RNNs suffer from long term dependencies when the cardinality of set grows bigger. The decoding process might stop early if the current hidden state cannot find any object and outputs the termination token. This would cause the remaining classes not to be predicted and lower recall metric. Transformers can be used to avoid the long term dependency problem exploiting their selfattention modules that process sequential data simultaneously. Consequently, we propose a novel transformer model for multi-label image classification which surpasses the state-of-the-art results by a large margin.
In the second part of thesis, we focus on two fashion-specific problems. Main product detection is the task of associating image parts with the fashion product that is being sold, generally using associated textual metadata (product title or description). Normally, in fashion e-commerces, products are represented by multiple images where a person wears the product along with other fashion items. If all the fashion items in the images are marked with bounding boxes, we can use the textual metadata to decide which item is the main product. The initial work treated each of these images independently, discarding the fact that they all belong to the same product. In this thesis, we represent the bounding boxes from all the images as nodes in a fully connected graph. This allows the algorithm to learn relations between the nodes during training and take the entire context into account for the final decision. Our algorithm results in a significant improvement of the state-ofthe-art.
Moreover, we address the problem of color naming for multicolored fashion items, which is a challenging task due to the external factors such as illumination changes or objects that act as clutter. In the context of multi-label classification, the vaguely defined lines between the classes in the color space cause ambiguity. For example, a shade of blue which is very close to green might cause the model to incorrectly predict the color blue and green at the same time. Based on this, models trained for color naming are expected to recognize the colors and their quantities in both single colored and multicolored fashion items. Therefore, in this thesis, we propose a novel architecture with an additional head that explicitly estimates the number of colors in fashion items. This removes the ambiguity problem and results in better color naming performance.

Abstract: The transcription of sheet music into some machine-readable format can be carried out manually. However, the complexity of music notation inevitably leads to burdensome software for music score editing, which makes the whole process
very time-consuming and prone to errors. Consequently, automatic transcription
systems for musical documents represent interesting tools.
Document analysis is the subject that deals with the extraction and processing
of documents through image and pattern recognition. It is a branch of computer
vision. Taking music scores as source, the field devoted to address this task is
known as Optical Music Recognition (OMR). Typically, an OMR system takes an
image of a music score and automatically extracts its content into some symbolic
structure such as MEI or MusicXML.
In this dissertation, we have investigated different methods for recognizing a
single staff section (e.g. scores for violin, flute, etc.), much in the same way as most text recognition research focuses on recognizing words appearing in a given line image. These methods are based in two different methodologies. On the one hand, we present two methods based on Recurrent Neural Networks, in particular, the
Long Short-Term Memory Neural Network. On the other hand, a method based on Sequence to Sequence models is detailed.
Music context is needed to improve the OMR results, just like language models
and dictionaries help in handwriting recognition. For example, syntactical rules
and grammars could be easily defined to cope with the ambiguities in the rhythm.
In music theory, for example, the time signature defines the amount of beats per
bar unit. Thus, in the second part of this dissertation, different methodologies
have been investigated to improve the OMR recognition. We have explored three
different methods: (a) a graphic tree-structure representation, Dendrograms, that
joins, at each level, its primitives following a set of rules, (b) the incorporation of Language Models to model the probability of a sequence of tokens, and (c) graph neural networks to analyze the music scores to avoid meaningless relationships between music primitives.
Finally, to train all these methodologies, and given the method-specificity of
the datasets in the literature, we have created four different music datasets. Two of them are synthetic with a modern or old handwritten appearance, whereas the
other two are real handwritten scores, being one of them modern and the other
old.

Abstract: Until recently, most computer vision models remained illiterate, largely ignoring the semantically rich and explicit information contained in scene text. Recent progress in scene text detection and recognition has recently allowed exploring its role in a diverse set of open computer vision problems, e.g. image classification, image-text retrieval, image captioning, and visual question answering to name a few. The explicit semantics of scene text closely requires specific modeling similar to language. However, scene text is a particular signal that has to be interpreted according to a comprehensive perspective that encapsulates all the visual cues in an image. Incorporating this information is a straightforward task for humans, but if we are unfamiliar with a language or scripture, achieving a complete world understanding is impossible (e.a. visiting a foreign country with a different alphabet). Despite the importance of scene text, modeling it requires considering the several ways in which scene text interacts with an image, processing and fusing an additional modality. In this thesis, we mainly focus
on two tasks, scene text-based fine-grained image classification, and cross-modal retrieval. In both studied tasks we identify existing limitations in current approaches and propose plausible solutions. Concretely, in each chapter: i) We define a compact way to embed scene text that generalizes to unseen words at training time while performing in real-time. ii) We incorporate the previously learned scene text embedding to create an image-level descriptor that overcomes optical character recognition (OCR) errors which is well-suited to the fine-grained image classification task. iii) We design a region-level reasoning network that learns the interaction through semantics among salient visual regions and scene text instances. iv) We employ scene text information in image-text matching and introduce the Scene Text Aware Cross-Modal retrieval StacMR task. We gather a dataset that incorporates scene text and design a model suited for the newly studied modality. v) We identify the drawbacks of current retrieval metrics in cross-modal retrieval. An image captioning metric is proposed as a way of better evaluating semantics in retrieved results. Ample experimentation shows that incorporating such semantics into a model yields better semantic results while
requiring significantly less data to converge.

Abstract: Human vision is restricted to the visual-optical spectrum. Machine vision is not.
Cameras sensitive to diverse infrared spectral bands can improve the capacities of
autonomous systems and provide a comprehensive view. Relevant scene content
can be made visible, particularly in situations when sensors of other modalities,
such as a visual-optical camera, require a source of illumination. As a result, increasing the level of automation not only avoids human errors but also reduces
machine-induced errors. Furthermore, multi-spectral sensor systems with infrared
imagery as one modality are a rich source of information and can conceivably
increase the robustness of many autonomous systems. Robotics, automobiles,
biometrics, security, surveillance, and the military are some examples of fields
that can profit from the use of infrared imagery in their respective applications.
Although multimodal spectral sensors have come a long way, there are still several
bottlenecks that prevent us from combining their output information and using
them as comprehensive images. The primary issue with infrared imaging is the lack
of potential benefits due to their cost influence on sensor resolution, which grows
exponentially with greater resolution. Due to the more costly sensor technology
required for their development, their resolutions are substantially lower than thoseof regular digital cameras.
This thesis aims to improve beyond-visible-spectrum machine vision by integrating multi-modal spectral sensors. The emphasis is on transforming the produced images to enhance their resolution to match expected human perception, bring the color representation close to human understanding of natural color, and improve machine vision application performance. This research focuses mainly on two tasks, image Colorization and Image Super resolution for both single- and cross-domain problems. We first start with an extensive review of the state of the art in both tasks, point out the shortcomings of existing approaches, and then present our solutions to address their limitations. Our solutions demonstrate that low-cost channel information (i.e., visible image) can be used to improve expensive channel
information (i.e., infrared image), resulting in images with higher quality and closer to human perception at a lower cost than a high-cost infrared camera.

Abstract: The curse of data labeling is a costly bottleneck in supervised deep learning, where large amounts of labeled data are needed to train intelligent systems. In onboard perception for autonomous driving, this cost corresponds to the labeling of raw data from sensors such as cameras, LiDARs, RADARs, etc. Therefore, synthetic data with automatically generated ground truth (labels) has aroused as a reliable alternative for training onboard perception models.
However, synthetic data commonly suffers from synth-to-real domain shift, i.e., models trained on the synthetic domain do not show their achievable accuracy when performing in the real world. This shift needs to be addressed by techniques falling in the realm of domain adaptation (DA).
The semi-supervised learning (SSL) paradigm can be followed to address DA. In this case, a model is trained using source data with labels (here synthetic) and leverages minimal knowledge from target data (here the real world) to generate pseudo-labels. These pseudo-labels help the training process to reduce the gap between the source and the target domains. In general, we can assume accessing both, pseudo-labels and a few amounts of human-provided labels for the target-domain data. However, the most interesting and challenging setting consists in assuming that we do not have human-provided labels at all. This setting is known as unsupervised domain adaptation (UDA). This PhD focuses on applying SSL to the UDA setting, for onboard visual tasks related to autonomous driving. We start by addressing the synth-to-real UDA problem on onboard vision-based object detection (pedestrians and cars), a critical task for autonomous driving and driving assistance. In particular, we propose to apply an SSL technique known as co-training, which we adapt to work with deep models that process a multi-modal input. The multi-modality consists of the visual appearance of the images (RGB) and their monocular depth estimation. The synthetic data we use as the source domain contains both, object bounding boxes and depth information. This prior knowledge is the
starting point for the co-training technique, which iteratively labels unlabeled real-world data and uses such pseudolabels (here bounding boxes with an assigned object class) to progressively improve the labeling results. Along this
process, two models collaborate to automatically label the images, in a way that one model compensates for the errors of the other, so avoiding error drift. While this automatic labeling process is done offline, the resulting pseudolabels can be used to train object detection models that must perform in real-time onboard a vehicle. We show that multi-modal co-training improves the labeling results compared to single-modal co-training, remaining competitive compared to human labeling.
Given the success of co-training in the context of object detection, we have also adapted this technique to a more crucial and challenging visual task, namely, onboard semantic segmentation. In fact, providing labels for a single image
can take from 30 to 90 minutes for a human labeler, depending on the content of the image. Thus, developing automatic labeling techniques for this visual task is of great interest to the automotive industry. In particular, the new co-training framework addresses synth-to-real UDA by an initial stage of self-training. Intermediate models arising from this stage are used to start the co-training procedure, for which we have elaborated an accurate collaboration policy between the two models performing the automatic labeling. Moreover, our co-training seamlessly leverages datasets from different synthetic domains. In addition, the co-training procedure is agnostic to the loss function used to train the semantic segmentation models which perform the automatic labeling. We achieve state-of-the-art results on publicly available benchmark datasets, again, remaining competitive compared to human labeling.
Finally, on the ground of our previous experience, we have designed and implemented a new SSL technique for UDA in the context of visual semantic segmentation. In this case, we mimic the labeling methodology followed by human labelers. In particular, rather than labeling full images at a time, categories of semantic classes are defined and only those are labeled in a labeling pass. In fact, different human labelers can become specialists in labeling different categories. Afterward, these per-category-labeled layers are combined to provide fully labeled images. Our technique is inspired by this methodology since we perform synth-to-real UDA per category, using the self-training stage previously developed as part of our co-training framework. The pseudo-labels obtained for each category are finally
fused to obtain fully automatically labeled images. In this context, we have also contributed to the development of a new photo-realistic synthetic dataset based on path-tracing rendering. Our new SSL technique seamlessly leverages publicly available synthetic datasets as well as this new one to obtain state-of-the-art results on synth-to-real UDA for semantic segmentation. We show that the new dataset allows us to reach better labeling accuracy than previously existing datasets, at the same time that it complements well them when combined. Moreover, we also show that the new human-inspired SSL technique outperforms co-training.

Abstract: Though they achieve great success, deep neural networks typically require a huge
amount of labeled data for training. However, collecting labeled data is often laborious and expensive. It would, therefore, be ideal if the knowledge obtained from label-rich datasets could be transferred to unlabeled data. However, deep networks are weak at generalizing to unseen domains, even when the differences are only subtle between the datasets. In real-world situations, a typical factor impairing the model generalization ability is the distribution shift between data from different domains, which is a long-standing problem usually termed as (unsupervised) domain adaptation.
A crucial requirement in the methodology of these domain adaptation methods is that they require access to source domain data during the adaptation process to the target domain. Accessibility to the source data of a trained source model is often impossible in real-world applications, for example, when deploying domain adaptation algorithms on mobile devices where the computational capacity is limited or in situations where data privacy rules limit access to the source domain data. Without access to the source domain data, existing methods suffer from inferior performance. Thus, in this thesis, we investigate domain adaptation without source data (termed as source-free domain adaptation) in multiple different scenarios that focus on image classification tasks.
We first study the source-free domain adaptation problem in a closed-set setting,
where the label space of different domains is identical. Only accessing the pretrained source model, we propose to address source-free domain adaptation from the perspective of unsupervised clustering. We achieve this based on nearest neighborhood clustering. In this way, we can transfer the challenging source-free domain adaptation task to a type of clustering problem. The final optimization objective is an upper bound containing only two simple terms, which can be explained as discriminability and diversity. We show that this allows us to relate several other methods in domain adaptation, unsupervised clustering and contrastive learning via the perspective of discriminability and diversity.