Refine
Departments, institutes and facilities
Document Type
- Article (3)
- Master's Thesis (2)
- Conference Object (1)
Language
- English (6)
Keywords
- deep learning (6) (remove)
Towards explaining deep learning networks to distinguish facial expressions of pain and emotions
(2018)
Deep learning networks are successfully used for object and face recognition in images and videos. In order to be able to apply such networks in practice, for example in hospitals as a pain recognition tool, the current procedures are only suitable to a limited extent. The advantage of deep learning methods is that they can learn complex non-linear relationships between raw data and target classes without limiting themselves to a set of hand-crafted features provided by humans. However, the disadvantage is that due to the complexity of these networks, it is not possible to interpret the knowledge that is stored inside the network. It is a black-box learning procedure. Explainable Artificial Intelligence (AI) approaches mitigate this problem by extracting explanations for decisions and representing them in a human-interpretable form. The aim of this paper is to investigate the explainable AI method Layer-wise Relevance Propagation (LRP) and apply it to explain how a deep learning network distinguishes facial expressions of pain from facial expressions of emotions such as happiness and disgust.
It is only a matter of time until autonomous vehicles become ubiquitous; however, human driving supervision will remain a necessity for decades. To assess the drive's ability to take control over the vehicle in critical scenarios, driver distractions can be monitored using wearable sensors or sensors that are embedded in the vehicle, such as video cameras. The types of driving distractions that can be sensed with various sensors is an open research question that this study attempts to answer. This study compared data from physiological sensors (palm electrodermal activity (pEDA), heart rate and breathing rate) and visual sensors (eye tracking, pupil diameter, nasal EDA (nEDA), emotional activation and facial action units (AUs)) for the detection of four types of distractions. The dataset was collected in a previous driving simulation study. The statistical tests showed that the most informative feature/modality for detecting driver distraction depends on the type of distraction, with emotional activation and AUs being the most promising. The experimental comparison of seven classical machine learning (ML) and seven end-to-end deep learning (DL) methods, which were evaluated on a separate test set of 10 subjects, showed that when classifying windows into distracted or not distracted, the highest F1-score of 79%; was realized by the extreme gradient boosting (XGB) classifier using 60-second windows of AUs as input. When classifying complete driving sessions, XGB's F1-score was 94%. The best-performing DL model was a spectro-temporal ResNet, which realized an F1-score of 75%; when classifying segments and an F1-score of 87%; when classifying complete driving sessions. Finally, this study identified and discussed problems, such as label jitter, scenario overfitting and unsatisfactory generalization performance, that may adversely affect related ML approaches.
Neural network based object detectors are able to automatize many difficult, tedious tasks. However, they are usually slow and/or require powerful hardware. One main reason is called Batch Normalization (BN) [1], which is an important method for building these detectors. Recent studies present a potential replacement called Self-normalizing Neural Network (SNN) [2], which at its core is a special activation function named Scaled Exponential Linear Unit (SELU). This replacement seems to have most of BNs benefits while requiring less computational power. Nonetheless, it is uncertain that SELU and neural network based detectors are compatible with one another. An evaluation of SELU incorporated networks would help clarify that uncertainty. Such evaluation is performed through series of tests on different neural networks. After the evaluation, it is concluded that, while indeed faster, SELU is still not as good as BN for building complex object detector networks.
A company's financial documents use tables along with text to organize the data containing key performance indicators (KPIs) (such as profit and loss) and a financial quantity linked to them. The KPI’s linked quantity in a table might not be equal to the similarly described KPI's quantity in a text. Auditors take substantial time to manually audit these financial mistakes and this process is called consistency checking. As compared to existing work, this paper attempts to automate this task with the help of transformer-based models. Furthermore, for consistency checking it is essential for the table's KPIs embeddings to encode the semantic knowledge of the KPIs and the structural knowledge of the table. Therefore, this paper proposes a pipeline that uses a tabular model to get the table's KPIs embeddings. The pipeline takes input table and text KPIs, generates their embeddings, and then checks whether these KPIs are identical. The pipeline is evaluated on the financial documents in the German language and a comparative analysis of the cell embeddings' quality from the three tabular models is also presented. From the evaluation results, the experiment that used the English-translated text and table KPIs and Tabbie model to generate table KPIs’ embeddings achieved an accuracy of 72.81% on the consistency checking task, outperforming the benchmark, and other tabular models.
The ability to finely segment different instances of various objects in an environment forms a critical tool in the perception tool-box of any autonomous agent. Traditionally instance segmentation is treated as a multi-label pixel-wise classification problem. This formulation has resulted in networks that are capable of producing high-quality instance masks but are extremely slow for real-world usage, especially on platforms with limited computational capabilities. This thesis investigates an alternate regression-based formulation of instance segmentation to achieve a good trade-off between mask precision and run-time. Particularly the instance masks are parameterized and a CNN is trained to regress to these parameters, analogous to bounding box regression performed by an object detection network.
In this investigation, the instance segmentation masks in the Cityscape dataset are approximated using irregular octagons and an existing object detector network (i.e., SqueezeDet) is modified to regresses to the parameters of these octagonal approximations. The resulting network is referred to as SqueezeDetOcta. At the image boundaries, object instances are only partially visible. Due to the convolutional nature of most object detection networks, special handling of the boundary adhering object instances is warranted. However, the current object detection techniques seem to be unaffected by this and handle all the object instances alike. To this end, this work proposes selectively learning only partial, untainted parameters of the bounding box approximation of the boundary adhering object instances. Anchor-based object detection networks like SqueezeDet and YOLOv2 have a discrepancy between the ground-truth encoding/decoding scheme and the coordinate space used for clustering, to generate the prior anchor shapes. To resolve this disagreement, this work proposes clustering in a space defined by two coordinate axes representing the natural log transformations of the width and height of the ground-truth bounding boxes.
When both SqueezeDet and SqueezeDetOcta were trained from scratch, SqueezeDetOcta lagged behind the SqueezeDet network by a massive ≈ 6.19 mAP. Further analysis revealed that the sparsity of the annotated data was the reason for this lackluster performance of the SqueezeDetOcta network. To mitigate this issue transfer-learning was used to fine-tune the SqueezeDetOcta network starting from the trained weights of the SqueezeDet network. When all the layers of the SqueezeDetOcta were fine-tuned, it outperformed the SqueezeDet network paired with logarithmically extracted anchors by ≈ 0.77 mAP. In addition to this, the forward pass latencies of both SqueezeDet and SqueezeDetOcta are close to ≈ 19ms. Boundary adhesion considerations, during training, resulted in an improvement of ≈ 2.62 mAP of the baseline SqueezeDet network. A SqueezeDet network paired with logarithmically extracted anchors improved the performance of the baseline SqueezeDet network by ≈ 1.85 mAP.
In summary, this work demonstrates that if given sufficient fine instance annotated data, an existing object detection network can be modified to predict much finer approximations (i.e., irregular octagons) of the instance annotations, whilst having the same forward pass latency as that of the bounding box predicting network. The results justify the merits of logarithmically extracted anchors to boost the performance of any anchor-based object detection network. The results also showed that the special handling of image boundary adhering object instances produces more performant object detectors.