Abstract: Word-spotting methods based on the Bag-ofVisual-Words framework have demonstrated a good retrieval performance even when used in a completely unsupervised manner. Although unsupervised approaches are suitable for
large document collections due to the cost of acquiring labeled data, these methods also present some drawbacks. For instance, having to train a suitable “codebook” for a certain dataset has a high computational cost. Therefore, in
this paper we present a database agnostic codebook which is trained from synthetic data. The aim of the proposed approach is to generate a codebook where the only information required is the type of script used in the document. The use of synthetic data also allows to easily incorporate semantic
information in the codebook generation. So, the proposed method is able to determine which set of codewords have a semantic representation of the descriptor feature space. Experimental results show that the resulting codebook attains a state-of-the-art performance while having a more compact representation.

Abstract: In this paper, we propose a precise synthetic ground truth dataset to study the problem of detection of the shadows cast by static objects in outdoor environments during extended periods of time (days). For our dataset, we have created a virtual scenario using a rendering software. To increase the realism of the simulated environment, we have defined the scenario in a precise geographical location. In our dataset the sun is by far the main illumination source. The sun position during the simulation time takes into consideration factors related to the geographical location, such as the latitude, longitude, elevation above sea level, and precise image capturing day and time. In our simulation the camera remains fixed. The dataset consists of seven days of simulation, from 10:00am to 5:00pm. Images are captured every 10 seconds. The shadows' ground truth is automatically computed by the rendering software.

Abstract: This paper presents a complete pipeline to perform deep learning-based instance segmentation of different types of grains (e.g., corn, sunflower, soybeans, lentils, chickpeas, mote, and beans). The proposed approach consists of using synthesized image datasets for the training process, which are easily generated according to the category of the instance to be segmented. The synthesized imaging process allows generating a large set of well-annotated grain samples with high variability—as large and high as the user requires. Instance segmentation is performed through a popular deep learning based approach, the Mask R-CNN architecture, but any learning-based instance segmentation approach can be considered. Results obtained by the proposed pipeline show that the strategy of using synthesized image datasets for training instance segmentation helps to avoid the time-consuming image annotation stage, as well as to achieve higher intersection over union and average precision performances. Results obtained with different varieties of grains are shown, as well as comparisons with manually annotated images, showing both the simplicity of the process and the improvements in the performance.

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: Action recognition in videos is a challenging task due to the complexity of the spatio-temporal patterns to model and the difficulty to acquire and learn on large quantities of video data. Deep learning, although a breakthrough for image classification and showing promise for videos, has still not clearly superseded action recognition methods using hand-crafted features, even when training on massive datasets. In this paper, we introduce hybrid video classification architectures based on carefully designed unsupervised representations of hand-crafted spatio-temporal features classified by supervised deep networks. As we show in our experiments on five popular benchmarks for action recognition, our hybrid model combines the best of both worlds: it is data efficient (trained on 150 to 10000 short clips) and yet improves significantly on the state of the art, including recent deep models trained on millions of manually labelled images and videos.

Abstract: In this paper, we address the problem of symbol spotting in technical document images applied to scanned and vectorized line drawings. Like any information spotting architecture, our approach has two components. First, symbols are decomposed in primitives which are compactly represented and second a primitive indexing structure aims to efficiently retrieve similar primitives. Primitives are encoded in terms of attributed strings representing closed regions. Similar strings are clustered in a lookup table so that the set median strings act as indexing keys. A voting scheme formulates hypothesis in certain locations of the line drawing image where there is a high presence of regions similar to the queried ones, and therefore, a high probability to find the queried graphical symbol. The proposed approach is illustrated in a framework consisting in spotting furniture symbols in architectural drawings. It has been proved to work even in the presence of noise and distortion introduced by the scanning and raster-to-vector processes.