China
Africa
Explore chapters and articles related to this topic
Segmentation Evaluation and Comparison
Published in Yu-Jin Zhang, A Selection of Image Analysis Techniques, 2023
An evaluation of six semi-automatic (cranio-maxillofacial surgery) image segmentation algorithms (with open-source codes) on three platforms has been reported (Wallner et al. 2019). The study and procedure have or share some representative characteristics for current image segmentation evaluation:Supervised evaluation: The quality and accuracy of segmentation methods was assessed by comparisons made between the segmentation algorithms and the ground truth segmentations. The ground truth results are often obtained with the help of human experts. For example, in this task, the ground truth of the anatomy structure is performed by clinical experts.Limited to a specified application domain: This is in contrast to early works that only consider the segmentation algorithms themselves while not consider the images and/or objects to be segmented. In this work, the medical (cranio-maxillofacial surgery) images are considered and the lower jaw (mandible) is chosen as the object (anatomical structure). More example can be found, for instance, specified to blood vessel (Moccia et al. 2018), high spatial resolution remote sensing image (Chen et al. 2018), or the histologic structures in the kidney cortex images (Jayapandian et al. 2021).
Audit of Artificial Intelligence Algorithms and Its Impact in Relieving Shortage of Specialist Doctors
Published in Sandeep Reddy, Artificial Intelligence, 2020
Vidur Mahajan, Vasanth Venugopal
Once a test dataset with corresponding ground truth has been assembled, it is important to determine the mix of cases required to aptly validate the AI algorithm. Generally, there are two types of algorithmic errors one is looking for – false positives and false negatives. False positives are cases that the AI calls out to be positive, but they in fact are negative (for the finding under question). To check for algorithms false-positive rate, it is important to have a dataset that comprises heavily of cases without many positives – this gives the validator a chance to see how frequently the AI calls a truly negative case as positive and is only possible when there are a high number of negatives. The same is true for false negatives, which are cases where in fact there is a finding, but the algorithm misses it. For this, a dataset comprising mainly of positive cases is required to determine whether (and to what extent) AI misses’ positive cases. This is especially important in today’s scenario when AI is being pegged as a tool for either triaging (Johnson et al., 2019) or for automatically identifying normal cases (Irvin et al., 2019; Singh et al., 2018) with a high degree of confidence.
Ground Truth Data for Remote Sensing Image Classification
Published in Anil Kumar, Priyadarshi Upadhyay, A. Senthil Kumar, Fuzzy Machine Learning Algorithms for Remote Sensing Image Classification, 2020
Anil Kumar, Priyadarshi Upadhyay, A. Senthil Kumar
Ground truth data as training data is used in supervised classification of an image. In the supervised classification approach, the ground truth data is used as training data. It is used to calculate the statistical parameters while using the statistical based classifier. The samples are generally collected through a combination of field work, maps, and field personal experience. The field sample locations are called training sites. In remote sensing classification, the spectral properties of training sites for each class are used to calculate the statistical or weight parameters. Some of the ground truth samples are used as testing data for the assessment of accuracy of the classified thematic maps since different classification algorithms may have varying percentages of errors for same classification data. Thus, it is helpful to identify the best classification algorithms with a given number of classes, while providing the least amount of error.
Regression-based detection of missing boundaries in multiphase polycrystalline microstructures
Published in Philosophical Magazine Letters, 2023
Manoj Prabakar, Prince Gideon Kubendran Amos
The training of the current regression-based detection approach involves introducing the ground truth by manually labelling of the objects. Put differently, the model to detect missing boundaries is trained by manually inserting bounding boxes around the corresponding defects in the micrographs. The ground truth defined in the form of the manually introduced labels, besides training, aides in refining the model during its validation. The performance of the model during testing and validation is ascertained through appropriate metrics and are graphically represented in Figure 2. The losses which indicate the deviation of the prediction from the ground truth are illustrated in Figure 2a. While the box loss quantifies the disparity between the parameters of the manual labels and the predicted bounding boxes, the difference in the confidences indicating the presence of the missing boundaries is expressed as the object loss. For a batch of four micrographs, Figure 2a shows that both box and object losses become increasingly negligible, as the number of epochs reach 1000. The minimal losses affirm the accuracy of the model in detecting missing boundaries. Other hyperparameters, besides epoch, are tuned to enhance the performance of the technique which includes learning rate, weight decay and momentum assuming values of 0.01, 0.0005 and 0.937, respectively.
A Unified Neuro-Fuzzy Framework to Assess the User Credibility on Twitter
Published in IETE Journal of Research, 2023
K. Laila, P. Jayashree, V. Vinuvarsidh
Yet another issue relates to the detection of bots. Rodríguez-Ruiz et al. [23] proposed a one-class classification mechanism that emphasized the behaviour of the bot creators. TwitterBot+, a framework [24] for detecting real-time bots solely relied on language-dependent statistical features extracted out of a single tweet. Bot detection approaches were categorized into crowd-sourcing, graph-based, and feature-based for the capture of users’ behaviour. The social bot detection approach [25,26] applied the pattern-based classification mechanism with the extension of the account age and the sentiment analysis. Acquisition of ground-truth datasets is time-consuming and needs human expertise. Therefore, labelled datasets were used for resolution of the challenges. However, predicting the type of user account using a single observation is under-represented.
Assessment of deep learning pose estimates for sports collision tracking
Published in Journal of Sports Sciences, 2022
Richard Blythman, Manan Saxena, Gregory J. Tierney, Chris Richter, Aljosa Smolic, Ciaran Simms
Differences between skeleton models of various motion capture systems and skeleton fitting procedures have long been studied (Leboeuf et al., 2019). Several works (Hicks & Richards, 2005; Leardini et al., 1999) have found the Davis equations (Davis et al., 1991) used in the original CGM for estimation of hip joint centres to be inadequate. While functional calibration methods were subsequently suggested as superior to the use of predictive equations, more recent analyses have found the issue to be with the specific equations used rather than the approach itself (Sangeux et al., 2011, 2014). These studies found the Harrington equations (Harrington et al., 2007) gave better estimates than the Davis equations. Hence, the Harrington equations are used in the current study. The configuration of the fitted skeleton model is shown in the schematic of Figure 1(b). This is used as the ground truth for evaluating the deep learning model.