Abstract: Understanding the content of the images is one of the greatest challenges of computer vision. Recognition of objects appearing in images, identifying and interpreting their actions are the main purposes of Image Understanding. This thesis seeks to identify what is present in a picture by categorizing and locating all the objects in the scene.
Images are composed by pixels, and one possibility consists of assigning to each pixel an object category, which is commonly known as semantic segmentation. By incorporating information as a contextual cue, we are able to resolve the ambiguity within categories at the pixel-level. We propose three levels of scale in order to resolve such ambiguity.
Another possibility to represent the objects is the object detection task. In this case, the aim is to recognize and localize the whole object by accurately placing a bounding box around it. We present two new approaches. The first one is focused on improving the object representation of deformable part models with the concept of factorized appearances. The second approach addresses the issue of reducing the computational cost for multi-class recognition. The results given have been validated on several commonly used datasets, reaching international recognition and state-of-the-art within the field

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: 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: Motivated by the success of powerful while expensive techniques to recognize words in a holistic way (Goel et al., 2013; Almazán et al., 2014; Jaderberg et al., 2016) object proposals techniques emerge as an alternative to the traditional text detectors. In this paper we introduce a novel object proposals method that is specifically designed for text. We rely on a similarity based region grouping algorithm that generates a hierarchy of word hypotheses. Over the nodes of this hierarchy it is possible to apply a holistic word recognition method in an efficient way.
Our experiments demonstrate that the presented method is superior in its ability of producing good quality word proposals when compared with class-independent algorithms. We show impressive recall rates with a few thousand proposals in different standard benchmarks, including focused or incidental text datasets, and multi-language scenarios. Moreover, the combination of our object proposals with existing whole-word recognizers (Almazán et al., 2014; Jaderberg et al., 2016) shows competitive performance in end-to-end word spotting, and, in some benchmarks, outperforms previously published results. Concretely, in the challenging ICDAR2015 Incidental Text dataset, we overcome in more than 10% F-score the best-performing method in the last ICDAR Robust Reading Competition (Karatzas, 2015). Source code of the complete end-to-end system is available at https://github.com/lluisgomez/TextProposals.

Abstract: Typography and layout lead to the hierarchical organisation of text in words, text lines, paragraphs. This inherent structure is a key property of text in any script and language, which has nonetheless been minimally leveraged by existing text detection methods. This paper addresses the problem of text
segmentation in natural scenes from a hierarchical perspective.
Contrary to existing methods, we make explicit use of text structure, aiming directly to the detection of region groupings corresponding to text within a hierarchy produced by an agglomerative similarity clustering process over individual regions. We propose an optimal way to construct such an hierarchy introducing a feature space designed to produce text group hypotheses with
high recall and a novel stopping rule combining a discriminative classifier and a probabilistic measure of group meaningfulness based in perceptual organization. Results obtained over four standard datasets, covering text in variable orientations and different languages, demonstrate that our algorithm, while being trained in a single mixed dataset, outperforms state of the art
methods in unconstrained scenarios.