Abstract: Image generation is arguably one of the most attractive, compelling, and challenging tasks in computer vision. Among the methods which perform image generation, generative adversarial networks (GANs) play a key role. The most common image generation models based on GANs can be divided into two main approaches. The first one, called simply image generation takes random noise as an input and synthesizes an image which follows the same distribution as the images in the training set. The second class, which is called image-to-image translation, aims to map an image from a source domain to one that is indistinguishable from those in the target domain. Image-to-image translation methods can further be divided into paired and unpaired image-to-image translation based on whether they require paired data or not. In this thesis, we aim to address some challenges of both image generation and image-to-image generation.GANs highly rely upon having access to vast quantities of data, and fail to generate realistic images from random noise when applied to domains with few images. To address this problem, we aim to transfer knowledge from a model trained on a large dataset (source domain) to the one learned on limited data (target domain). We find that both GANs andconditional GANs can benefit from models trained on large datasets. Our experiments show that transferring the discriminator is more important than the generator. Using both the generator and discriminator results in the best performance. We found, however, that this method suffers from overfitting, since we update all parameters to adapt to the target data. We propose a novel architecture, which is tailored to address knowledge transfer to very small target domains. Our approach effectively exploreswhich part of the latent space is more related to the target domain. Additionally, the proposed method is able to transfer knowledge from multiple pretrained GANs. Although image-to-image translation has achieved outstanding performance, it still facesseveral problems. First, for translation between complex domains (such as translations between different modalities) image-to-image translation methods require paired data. We show that when only some of the pairwise translations have been seen (i.e. during training), we can infer the remaining unseen translations (where training pairs are not available). We propose a new approach where we align multiple encoders and decoders in such a way that the desired translation can be obtained by simply cascadingthe source encoder and the target decoder, even when they have not interacted during the training stage (i.e. unseen). Second, we address the issue of bias in image-to-image translation. Biased datasets unavoidably contain undesired changes, which are dueto the fact that the target dataset has a particular underlying visual distribution. We use carefully designed semantic constraints to reduce the effects of the bias. The semantic constraint aims to enforce the preservation of desired image properties. Finally, current approaches fail to generate diverse outputs or perform scalable image transfer in a single model. To alleviate this problem, we propose a scalable and diverse image-to-image translation. We employ random noise to control the diversity. The scalabitlity is determined by conditioning the domain label.computer vision, deep learning, imitation learning, adversarial generative networks, image generation, image-to-image translation.

Abstract: Manually annotating images to develop vision models has been a major bottleneck
since computer vision and machine learning started to walk together. This has
been more evident since computer vision falls on the shoulders of data-hungry
deep learning techniques. When addressing on-board perception for autonomous
driving, the curse of data annotation is exacerbated due to the use of additional
sensors such as LiDAR. Therefore, any approach aiming at reducing such a timeconsuming and costly work is of high interest for addressing autonomous driving
and, in fact, for any application requiring some sort of artificial perception. In the
last decade, it has been shown that leveraging from synthetic data is a paradigm
worth to pursue in order to minimizing manual data annotation. The reason is
that the automatic process of generating synthetic data can also produce different
types of associated annotations (e.g. object bounding boxes for synthetic images
and LiDAR pointclouds, pixel/point-wise semantic information, etc.). Directly
using synthetic data for training deep perception models may not be the definitive
solution in all circumstances since it can appear a synth-to-real domain shift. In
this context, this work focuses on leveraging synthetic data to alleviate manual
annotation for three perception tasks related to driving assistance and autonomous
driving. In all cases, we assume the use of deep convolutional neural networks
(CNNs) to develop our perception models.
The first task addresses traffic sign recognition (TSR), a kind of multi-class
classification problem. We assume that the number of sign classes to be recognized
must be suddenly increased without having annotated samples to perform the
corresponding TSR CNN re-training. We show that leveraging synthetic samples of
such new classes and transforming them by a generative adversarial network (GAN)
trained on the known classes (i.e. without using samples from the new classes), it is
possible to re-train the TSR CNN to properly classify all the signs for a ∼ 1/4 ratio of
new/known sign classes. The second task addresses on-board 2D object detection,
focusing on vehicles and pedestrians. In this case, we assume that we receive a set
of images without the annotations required to train an object detector, i.e. without
object bounding boxes. Therefore, our goal is to self-annotate these images so
that they can later be used to train the desired object detector. In order to reach
this goal, we leverage from synthetic data and propose a semi-supervised learning
approach based on the co-training idea. In fact, we use a GAN to reduce the synthto-real domain shift before applying co-training. Our quantitative results show
that co-training and GAN-based image-to-image translation complement each
other up to allow the training of object detectors without manual annotation, and still almost reaching the upper-bound performances of the detectors trained from
human annotations. While in previous tasks we focus on vision-based perception,
the third task we address focuses on LiDAR pointclouds. Our initial goal was to
develop a 3D object detector trained on synthetic LiDAR-style pointclouds. While
for images we may expect synth/real-to-real domain shift due to differences in
their appearance (e.g. when source and target images come from different camera
sensors), we did not expect so for LiDAR pointclouds since these active sensors
factor out appearance and provide sampled shapes. However, in practice, we have
seen that it can be domain shift even among real-world LiDAR pointclouds. Factors
such as the sampling parameters of the LiDARs, the sensor suite configuration onboard the ego-vehicle, and the human annotation of 3D bounding boxes, do induce
a domain shift. We show it through comprehensive experiments with different
publicly available datasets and 3D detectors. This redirected our goal towards the
design of a GAN for pointcloud-to-pointcloud translation, a relatively unexplored
topic.
Finally, it is worth to mention that all the synthetic datasets used for these three
tasks, have been designed and generated in the context of this PhD work and will
be publicly released. Overall, we think this PhD presents several steps forward to
encourage leveraging synthetic data for developing deep perception models in the
field of driving assistance and autonomous driving.

Abstract: Machine learning experimentation under controlled scenarios and standard datasets is necessary to compare algorithms performance by evaluating all of them in the same setup. However, experimentation on how those algorithms perform on unconstrained data and applied tasks to solve real world problems is also a must to ascertain how that research can contribute to our society.
In this dissertation we experiment with the latest computer vision and natural language processing algorithms applying them to multimodal scene interpretation. Particularly, we research on how image and text understanding can be jointly exploited to address real world problems, focusing on learning from Social Media data.
We address several tasks that involve image and textual information, discuss their characteristics and offer our experimentation conclusions. First, we work on detection of scene text in images. Then, we work with Social Media posts, exploiting the captions associated to images as supervision to learn visual features, which we apply to multimodal semantic image retrieval. Subsequently, we work with geolocated Social Media images with associated tags, experimenting on how to use the tags as supervision, on location sensitive image retrieval and on exploiting location information for image tagging. Finally, we work on a specific classification problem of Social Media publications consisting on an image and a text: Multimodal hate speech classification.

Abstract: Computer vision has gone through considerable changes in the last decade as neural networks have come into common use. As available computational capabilities have grown, neural networks have achieved breakthroughs in many computer vision tasks, and have even surpassed human performance in others. With accuracy being so high, focus has shifted to other issues and challenges. One research direction that saw a notable increase in interest is on lifelong learning systems. Such systems should be capable of efficiently performing tasks, identifying and learning new ones, and should moreover be able to deploy smaller versions of themselves which are experts on specific tasks. In this thesis, we contribute to research on lifelong learning and address the compression and adaptation of networks to small target domains, the incremental learning of networks faced with a variety of tasks, and finally the detection of out-of-distribution samples at inference time.
We explore how knowledge can be transferred from large pretrained models to more task-specific networks capable of running on smaller devices by extracting the most relevant information. Using a pretrained model provides more robust representations and a more stable initialization when learning a smaller task, which leads to higher performance and is known as domain adaptation. However, those models are too large for certain applications that need to be deployed on devices with limited memory and computational capacity. In this thesis we show that, after performing domain adaptation, some learned activations barely contribute to the predictions of the model. Therefore, we propose to apply network compression based on low-rank matrix decomposition using the activation statistics. This results in a significant reduction of the model size and the computational cost.
Like human intelligence, machine intelligence aims to have the ability to learn and remember knowledge. However, when a trained neural network is presented with learning a new task, it ends up forgetting previous ones. This is known as catastrophic forgetting and its avoidance is studied in continual learning. The work presented in this thesis extensively surveys continual learning techniques and presents an approach to avoid catastrophic forgetting in sequential task learning scenarios. Our technique is based on using ternary masks in order to update a network to new tasks, reusing the knowledge of previous ones while not forgetting anything about them. In contrast to earlier work, our masks are applied to the activations of each layer instead of the weights. This considerably reduces the number of parameters to be added for each new task. Furthermore, the analysis on a wide range of work on incremental learning without access to the task-ID, provides insight on current state-of-the-art approaches that focus on avoiding catastrophic forgetting by using regularization, rehearsal of previous tasks from a small memory, or compensating the task-recency bias.
Neural networks trained with a cross-entropy loss force the outputs of the model to tend toward a one-hot encoded vector. This leads to models being too overly confident when presented with images or classes that were not present in the training distribution. The capacity of a system to be aware of the boundaries of the learned tasks and identify anomalies or classes which have not been learned yet is key to lifelong learning and autonomous systems. In this thesis, we present a metric learning approach to out-of-distribution detection that learns the task at hand on an embedding space.

Abstract: Sectors as fintech, legaltech or insurance process an inflow of millions of forms, invoices, id documents, claims or similar every day. Together with these, historical archives provide gigantic amounts of digitized documents containing useful information that needs to be stored in machine encoded text with a meaningful structure. This procedure, known as information extraction (IE) comprises the steps of localizing and recognizing text, identifying named entities contained in it and optionally finding relationships among its elements. In this work we explore multi-task neural models at image and graph level to solve all steps in a unified way. While doing so we find benefits and limitations of these end-to-end approaches in comparison with sequential separate methods. More specifically, we first propose a method to produce textual as well as semantic labels with a unified model from handwritten text line images. We do so with the use of a convolutional recurrent neural model trained with connectionist temporal classification to predict the textual as well as semantic information encoded in the images. Secondly, motivated by the success of this approach we investigate the unification of the localization and recognition tasks of handwritten text in full pages with an end-to-end model, observing benefits in doing so. Having two models that tackle information extraction subsequent task pairs in an end-to-end to end manner, we lastly contribute with a method to put them all together in a single neural network to solve the whole information extraction pipeline in a unified way. Doing so we observe some benefits and some limitations in the approach, suggesting that in certain cases it is beneficial to train specialized models that excel at a single challenging task of the information extraction process, as it can be the recognition of named entities or the extraction of relationships between them. For this reason we lastly study the use of the recently arrived graph neural network architectures for the semantic tasks of the information extraction process, which are recognition of named entities and relation extraction, achieving promising results on the relation extraction part.

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: 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.