Deep neural networks have recently led to promising results for the task of multiple sound source localization. Yet, they require a lot of training data to cover a variety of acoustic conditions and micro-phone array layouts. One can leverage acoustic simulators to inexpensively generate labeled training data. However, models trained on synthetic data tend to perform poorly with real-world recordings due to the domain mismatch. Moreover, learning for different microphone array layouts makes the task more complicated due to the infinite number of possible layouts. We propose to use adversarial learning methods to close the gap between synthetic and real do-mains. Our novel ensemble-discrimination method significantly improves the localization performance without requiring any label from the real data. Furthermore, we propose a novel explicit transformation layer to be embedded in the localization architecture. It enables the model to be trained with data from specific microphone array layouts while generalizing well to unseen layouts during inference.

In this paper, we study the problem of unifying knowledge from a set of classifiers with different architectures and target classes into a single classifier, given only a generic set of unlabelled data. We call this problem Unifying Heterogeneous Classifiers (UHC). This problem is motivated by scenarios where data is collected from multiple sources, but the sources cannot share their data, e.g., due to privacy concerns, and only privately trained models can be shared. In addition, each source may not be able to gather data to train all classes due to data availability at each source, and may not be able to train the same classification model due to different computational resources. To tackle this problem, we propose a generalisation of knowledge distillation to merge HCs. We derive a probabilistic relation between the outputs of HCs and the probability over all classes. Based on this relation, we propose two classes of methods based on cross-entropy minimisation and matrix factorisation, which allow us to estimate soft labels over all classes from unlabelled samples and use them in lieu of ground truth labels to train a unified classifier. Our extensive experiments on ImageNet, LSUN, and Places365 datasets show that our approaches significantly outperform a naive extension of distillation and can achieve almost the same accuracy as classifiers that are trained in a centralised, supervised manner.

Image-based sports analytics enable automatic retrieval of key events in a game to speed up the analytics process for human experts. However, most existing methods focus on structured television broadcast video datasets with a straight and fixed camera having minimum variability in the capturing pose. In this paper, we study the case of event detection in sports videos for unstructured environments with arbitrary camera angles. The transition from structured to unstructured video analysis produces multiple challenges that we address in our paper. Specifically, we identify and solve two major problems: unsupervised identification of players in an unstructured setting and generalization of the trained models to pose variations due to arbitrary shooting angles. For the first problem, we propose a temporal feature aggregation algorithm using person re-identification features to obtain high player retrieval precision by boosting a weak heuristic scoring method. Additionally, we propose a data augmentation technique, based on multi-modal image translation model, to reduce bias in the appearance of training samples. Experimental evaluations show that our proposed method improves precision for player retrieval from 0.78 to 0.86 for obliquely angled videos. Additionally, we obtain an improvement in F1 score for rally detection in table tennis videos from 0.79 in case of global frame-level features to 0.89 using our proposed player-level features. Please see the supplementary video submission at https://ibm.biz/BdzeZA.

In this paper, we propose novel deep learning based algorithms for multiple sound source localization. Specifically, we aim to find the 2D Cartesian coordinates of multiple sound sources in an enclosed environment by using multiple microphone arrays. To this end, we use an encoding-decoding architecture and propose two improvements on it to accomplish the task. In addition, we also propose two novel localization representations which increase the accuracy. Lastly, new metrics are developed relying on resolution-based multiple source association which enables us to evaluate and compare different localization approaches. We tested our method on both synthetic and real world data. The results show that our method improves upon the previous baseline approach for this problem.

This paper is a contribution towards interpretability of the deep learning models in different applications of time-series. We propose a temporal attention layer that is capable of selecting the relevant information to perform various tasks, including data completion, key-frame detection and classification. The method uses the whole input sequence to calculate an attention value for each time step. This results in more focused attention values and more plausible visualisation than previous methods. We apply the proposed method to three different tasks. Experimental results show that the proposed network produces comparable results to a state of the art. In addition, the network provides better interpretability of the decision, that is, it generates more significant attention weight to related frames compared to similar techniques attempted in the past.

Spatio-temporal anomaly detection by unsupervised learning have applications in a wide range of practical settings. In this paper we present a surveillance system for industrial robots using a monocular camera. We propose a new unsupervised learning method to train a deep feature extractor from unlabeled images. Without any data augmentation, the algorithm co-learns the network parameters on different pseudo-classes simultaneously to create unbiased feature representation. Combining the learned features with a prediction system, we can detect irregularities in high dimensional data feed (e.g. video of a robot performing pick and place task). The results show how the proposed approach can detect previously unseen anomalies in the robot surveillance video. Although the technique is not designed for classification, we show the use of the learned features in a more traditional classification application for CIFAR-10 dataset.

The goal of this work is to teach a robot to regrasp an object using knowledge obtained from human demonstration. This paper presents a task model that represents a human regrasping movement. The task model is based on the topological information and comprised of four task primitives. Human regrasping movement is recognised and represented as a sequence of these task primitives by the proposed recognition algorithm. The proposed method then maps each task primitive to the target robot hand using knowledge obtained from human demonstration. The experimental result verified the proposed task model by executing the regrasping movement on the real robot hand.

This paper introduces a novel method to teach a robot to regrasp an object based on the Programming by Demonstration paradigm. In this paradigm, a robot observes a human performing a regrasping task via various sensors to recognise crucial information in order to reproduce the task using its own hand. The main contribution is in the proposal of a representation technique that can analyse a human regrasping movement and reproduce this movement in a robot hand. The technique is based on tangle topology where both hand and manipulated object are considered as strands. This allows a regrasping movement to be considered as an alteration of the tangle relationship between the strands (hand and the object) over time. Human regrasping movements are analysed and reproduced in multi-fingered robot hands in a grasp simulation to demonstrate the efficiency of the proposed method.

This paper presents a novel method for robust dance motion structure detection. In the japanese folk dance domain, teachers created illustrations of dance poses. These poses characterize the most important movements of a dance. So far there is no simple and reliable extraction method which can extract all poses as shown in these drawings. We use these poses for the Task Model (TM) in the context of Learning from Observation (LFO). LFO which is a well known technique for successful human to robot motion mapping, consists of tasks (what to do) and skills (how to do). We propose a novel approach, to extract special motions from a dance, called turning motions useful for skill mapping in the LFO paradigm. Furthermore, we use a modified version of this approach, to detect all poses as shown in the drawings, called turning poses. To achieve this we observe both forearms at the same time and analyze their movement in different 2-D coordinate planes. We evaluate the parameters with and without a weighting function where we minimize acceleration, velocity and power. We successfully demonstrate this novel method using two very different japanese folk dances and discuss further implications of this work in respect to the LFO paradigm and dances of other domains.