Abstract: In this article we investigate the problem of human action recognition in static images. By action recognition we intend a class of problems which includes both action classification and action detection (i.e. simultaneous localization and classification). Bag-of-words image representations yield promising results for action classification, and deformable part models perform very well object detection. The representations for action recognition typically use only shape cues and ignore color information. Inspired by the recent success of color in image classification and object detection, we investigate the potential of color for action classification and detection in static images. We perform a comprehensive evaluation of color descriptors and fusion approaches for action recognition. Experiments were conducted on the three datasets most used for benchmarking action recognition in still images: Willow, PASCAL VOC 2010 and Stanford-40. Our experiments demonstrate that incorporating color information considerably improves recognition performance, and that a descriptor based on color names outperforms pure color descriptors. Our experiments demonstrate that late fusion of color and shape information outperforms other approaches on action recognition. Finally, we show that the different color–shape fusion approaches result in complementary information and combining them yields state-of-the-art performance for action classification.

Abstract: Semantic image segmentation is a central and challenging task in autonomous driving, addressed by training deep models. Since this training draws to a curse of human-based image labeling, using synthetic images with automatically generated labels together with unlabeled real-world images is a promising alternative. This implies to address an unsupervised domain adaptation (UDA) problem. In this paper, we propose a new co-training procedure for synth-to-real UDA of semantic
segmentation models. It consists of a self-training stage, which provides two domain-adapted models, and a model collaboration loop for the mutual improvement of these two models. These models are then used to provide the final semantic segmentation labels (pseudo-labels) for the real-world images. The overall
procedure treats the deep models as black boxes and drives their collaboration at the level of pseudo-labeled target images, i.e., neither modifying loss functions is required, nor explicit feature alignment. We test our proposal on standard synthetic and real-world datasets for on-board semantic segmentation. Our
procedure shows improvements ranging from ∼13 to ∼26 mIoU points over baselines, so establishing new state-of-the-art results.

Abstract: Top-performing computer vision models are powered by convolutional neural networks (CNNs). Training an accurate CNN highly depends on both the raw sensor data and their associated ground truth (GT). Collecting such GT is usually done through human labeling, which is time-consuming and does not scale as we wish. This data-labeling bottleneck may be intensified due to domain shifts among image sensors, which could force per-sensor data labeling. In this paper, we focus on the use of co-training, a semi-supervised learning (SSL) method, for obtaining self-labeled object bounding boxes (BBs), i.e., the GT to train deep object detectors. In particular, we assess the goodness of multi-modal co-training by relying on two different views of an image, namely, appearance (RGB) and estimated depth (D). Moreover, we compare appearance-based single-modal co-training with multi-modal. Our results suggest that in a standard SSL setting (no domain shift, a few human-labeled data) and under virtual-to-real domain shift (many virtual-world labeled data, no human-labeled data) multi-modal co-training outperforms single-modal. In the latter case, by performing GAN-based domain translation both co-training modalities are on par, at least when using an off-the-shelf depth estimation model not specifically trained on the translated images.

Abstract: Having artificial agents to autonomously produce human-like behaviour is one of the most ambitious original goals of Artificial Intelligence (AI) and remains an open problem nowadays. The imitation game originally proposed by Turing constitute a very effective method to prove the indistinguishability of an artificial agent. The behaviour of an agent is said to be indistinguishable from that of a human when observers (the so-called judges in the Turing test) cannot tell apart humans and non-human agents. Different environments, testing protocols, scopes and problem domains can be established to develop limited versions or variants of the original Turing test. In this paper we use a specific version of the Turing test, based on the international BotPrize competition, built in a First-Person Shooter video game, where both human players and non-player characters interact in complex virtual environments. Based on our past experience both in the BotPrize competition and other robotics and computer game AI applications we have developed three new more advanced controllers for believable agents: two based on a combination of the CERA–CRANIUM and SOAR cognitive architectures and other based on ADANN, a system for the automatic evolution and adaptation of artificial neural networks. These two new agents have been put to the test jointly with CCBot3, the winner of BotPrize 2010 competition (Arrabales et al., 2012), and have showed a significant improvement in the humanness ratio. Additionally, we have confronted all these bots to both First-person believability assessment (BotPrize original judging protocol) and Third-person believability assessment, demonstrating that the active involvement of the judge has a great impact in the recognition of human-like behaviour.