Dalia Ortiz Pablo, Sushruth Badri, Erik Norén, Christoph Nötzli
A major problem with using automated classification systems is that if they are not engineered correctly and with fairness considerations, they could be detrimental to certain populations. Furthermore, while engineers have developed cutting-edge technologies for image classification, there is still a gap in the application of these models in human heritage collections, where data sets usually consist of low-quality pictures of people with diverse ethnicity, gender, and age. In this work, we evaluate three bias mitigation techniques using two state-of-the-art neural networks, Xception and EfficientNet, for gender classification. Moreover, we explore the use of transfer learning using a fair data set to overcome the training data scarcity. We evaluated the effectiveness of the bias mitigation pipeline on a cultural heritage collection of photographs from the 19th and 20th centuries, and we used the FairFace data set for the transfer learning experiments. After the evaluation, we found that transfer learning is a good technique that allows better performance when working with a small data set. Moreover, the fairest classifier was found to be accomplished using transfer learning, threshold change, re-weighting and image augmentation as bias mitigation methods.
{"title":"Bias mitigation techniques in Image Classification: Fair Machine Learning in Human Heritage Collections","authors":"Dalia Ortiz Pablo, Sushruth Badri, Erik Norén, Christoph Nötzli","doi":"10.24132/jwscg.2023.6","DOIUrl":"https://doi.org/10.24132/jwscg.2023.6","url":null,"abstract":"A major problem with using automated classification systems is that if they are not engineered correctly and with fairness considerations, they could be detrimental to certain populations. Furthermore, while engineers have developed cutting-edge technologies for image classification, there is still a gap in the application of these models in human heritage collections, where data sets usually consist of low-quality pictures of people with diverse ethnicity, gender, and age. In this work, we evaluate three bias mitigation techniques using two state-of-the-art neural networks, Xception and EfficientNet, for gender classification. Moreover, we explore the use of transfer learning using a fair data set to overcome the training data scarcity. We evaluated the effectiveness of the bias mitigation pipeline on a cultural heritage collection of photographs from the 19th and 20th centuries, and we used the FairFace data set for the transfer learning experiments. After the evaluation, we found that transfer learning is a good technique that allows better performance when working with a small data set. Moreover, the fairest classifier was found to be accomplished using transfer learning, threshold change, re-weighting and image augmentation as bias mitigation methods.","PeriodicalId":39283,"journal":{"name":"Journal of WSCG","volume":"214 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135509301","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Vanessa Suessle, Mimi Arandjelovic, Ammie K. Kalan, Anthony Agbor, Christophe Boesch, Gregory Brazzola, Tobias Deschner, Paula Dieguez, Anne-Céline Granjon, Hjalmar Kuehl, Anja Landsmann, Juan Lapuente, Nuria Maldonado, Amelia Meier, Zuzana Rockaiova, Erin G. Wessling, Roman M. Wittig, Colleen T. Downs, Andreas Weinmann, Elke Hergenroether
The manual processing and analysis of videos from camera traps is time-consuming and includes several steps, ranging from the filtering of falsely triggered footage to identifying and re-identifying individuals. In this study, we developed a pipeline to automatically analyze videos from camera traps to identify individuals without requiring manual interaction. This pipeline applies to animal species with uniquely identifiable fur patterns and solitary behavior, such as leopards (Panthera pardus). We assumed that the same individual was seen throughout one triggered video sequence. With this assumption, multiple images could be assigned to an individual for the initial database filling without pre-labeling. The pipeline was based on well-established components from computer vision and deep learning, particularly convolutional neural networks (CNNs) and scale-invariant feature transform (SIFT) features. We augmented this basis by implementing additional components to substitute otherwise required human interactions. Based on the similarity between frames from the video material, clusters were formed that represented individuals bypassing the open set problem of the unknown total population. The pipeline was tested on a dataset of leopard videos collected by the Pan African Programme: The Cultured Chimpanzee (PanAf) and achieved a success rate of over 83% for correct matches between previously unknown individuals. The proposed pipeline can become a valuable tool for future conservation projects based on camera trap data, reducing the work of manual analysis for individual identification, when labeled data is unavailable.
{"title":"Automatic Individual Identification of Patterned Solitary Species Based on Unlabeled Video Data","authors":"Vanessa Suessle, Mimi Arandjelovic, Ammie K. Kalan, Anthony Agbor, Christophe Boesch, Gregory Brazzola, Tobias Deschner, Paula Dieguez, Anne-Céline Granjon, Hjalmar Kuehl, Anja Landsmann, Juan Lapuente, Nuria Maldonado, Amelia Meier, Zuzana Rockaiova, Erin G. Wessling, Roman M. Wittig, Colleen T. Downs, Andreas Weinmann, Elke Hergenroether","doi":"10.24132/jwscg.2023.1","DOIUrl":"https://doi.org/10.24132/jwscg.2023.1","url":null,"abstract":"The manual processing and analysis of videos from camera traps is time-consuming and includes several steps, ranging from the filtering of falsely triggered footage to identifying and re-identifying individuals. In this study, we developed a pipeline to automatically analyze videos from camera traps to identify individuals without requiring manual interaction. This pipeline applies to animal species with uniquely identifiable fur patterns and solitary behavior, such as leopards (Panthera pardus). We assumed that the same individual was seen throughout one triggered video sequence. With this assumption, multiple images could be assigned to an individual for the initial database filling without pre-labeling. The pipeline was based on well-established components from computer vision and deep learning, particularly convolutional neural networks (CNNs) and scale-invariant feature transform (SIFT) features. We augmented this basis by implementing additional components to substitute otherwise required human interactions. Based on the similarity between frames from the video material, clusters were formed that represented individuals bypassing the open set problem of the unknown total population. The pipeline was tested on a dataset of leopard videos collected by the Pan African Programme: The Cultured Chimpanzee (PanAf) and achieved a success rate of over 83% for correct matches between previously unknown individuals. The proposed pipeline can become a valuable tool for future conservation projects based on camera trap data, reducing the work of manual analysis for individual identification, when labeled data is unavailable.","PeriodicalId":39283,"journal":{"name":"Journal of WSCG","volume":"615 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135509293","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-01DOI: 10.24132/jwscg.2021.29.2
Jacob D. Hauenstein, Timothy S Newman
New methods for hypersurface (that is, 3-dimensional manifold) curvature determination in volumetric data areintroduced. One method is convolution-based. Another method is spline-based. Method accuracy is also analyzed,with that analysis involving comparison of the methods with each other as well as against two existing convolution-based methods. The accuracy analysis utilizes a novel framework that enables curvature determination methodaccuracy analysis via dynamically generated synthetic test datasets formed from continuous trivariate functions.Such functions enable accuracy analysis versus ground truth. The framework is also described here.
{"title":"New Methods and Novel Framework for HypersurfaceCurvature Determination and Analysis","authors":"Jacob D. Hauenstein, Timothy S Newman","doi":"10.24132/jwscg.2021.29.2","DOIUrl":"https://doi.org/10.24132/jwscg.2021.29.2","url":null,"abstract":"New methods for hypersurface (that is, 3-dimensional manifold) curvature determination in volumetric data areintroduced. One method is convolution-based. Another method is spline-based. Method accuracy is also analyzed,with that analysis involving comparison of the methods with each other as well as against two existing convolution-based methods. The accuracy analysis utilizes a novel framework that enables curvature determination methodaccuracy analysis via dynamically generated synthetic test datasets formed from continuous trivariate functions.Such functions enable accuracy analysis versus ground truth. The framework is also described here.","PeriodicalId":39283,"journal":{"name":"Journal of WSCG","volume":"56 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2021-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84392087","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-01DOI: 10.24132/jwscg.2021.29.1
Damia Fuentes Escote, S. Semwal
Using drones and augmented reality paradigm, new forms of interactive algorithms has been created and proposed.We start with a first person view interaction where the drone mimics the movement of one person’s head wearing aHMD so that movements of the head can be mapped to actions by the drones. We then provide two novel AR/VRapplications of drones to create something similar to third person view in 2D and 3D. To get started, our firstidea is to control a drone using head movements. The second application which we implemented is to provide animplementation where tangible platforms are used by the drone to react to the movements of the character. Finallyour third implementaton if to create and AR world using real outdoor scenery and asking a drone to mimic a thirdperson view combining the real scenery with a synthetic actor so that based on the synthetic actor movement thedrone changes its behavior correctly in the real-word trying to provide a synchronized view of the real and syntheticword. There are three novel ideas providing a new form of interactions which will improve with drones functionalityin future. Our implementation shows the feasibility of our idea as discussed in the paper.
{"title":"Drone interactions within the field of Augmented Reality","authors":"Damia Fuentes Escote, S. Semwal","doi":"10.24132/jwscg.2021.29.1","DOIUrl":"https://doi.org/10.24132/jwscg.2021.29.1","url":null,"abstract":"Using drones and augmented reality paradigm, new forms of interactive algorithms has been created and proposed.We start with a first person view interaction where the drone mimics the movement of one person’s head wearing aHMD so that movements of the head can be mapped to actions by the drones. We then provide two novel AR/VRapplications of drones to create something similar to third person view in 2D and 3D. To get started, our firstidea is to control a drone using head movements. The second application which we implemented is to provide animplementation where tangible platforms are used by the drone to react to the movements of the character. Finallyour third implementaton if to create and AR world using real outdoor scenery and asking a drone to mimic a thirdperson view combining the real scenery with a synthetic actor so that based on the synthetic actor movement thedrone changes its behavior correctly in the real-word trying to provide a synchronized view of the real and syntheticword. There are three novel ideas providing a new form of interactions which will improve with drones functionalityin future. Our implementation shows the feasibility of our idea as discussed in the paper.","PeriodicalId":39283,"journal":{"name":"Journal of WSCG","volume":"11 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2021-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"83001341","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-01-01DOI: 10.24132/jwscg.2021.29.4
Felippe T. Angelo, R. Voltoline, G. Gonçalves, Shin-Ting Wu
As the imaging technology and the understanding of neurological disease improve, a solid understanding of neu-roanatomy has become increasingly relevant. Neuroanatomy teaching includes the practice of cadaveric dissectionand neuroanatomy atlases consisting of images of a brain with its labeled structures. However, the natural inter-individual neuroanatomical variability cannot be taken into account. This work addresses the individual grossneuroanatomy atlas that could enrich medical students’ experiences with various individual variations in anatomi-cal landmarks and their spatial relationships. We propose to deform the CerebrA cortical atlas into the individualanatomical magnetic resonance imaging data to increase students’ opportunity to contact normal neuroanatomicalvariations in the early stages of studies. Besides, we include interactive queries on the labels/names of neu-roanatomical structures from an individual neuroanatomical atlas in a 3D space. An implementation on top ofSimpleITK library and VMTK-Neuro software is presented. We generated a series of surface and internal neu-roanatomy maps from 16 test volumes to attest to the potential of the proposed technique in brain labeling. Forthe age group between 10 to 75, there is evidence that the superficial cortical labeling is accurate with the visualassessment of the degree of concordance between the neuroanatomical and label boundaries.
{"title":"Interactive Individualized Neuroanatomy Labeling for Neuroanatomy Teaching","authors":"Felippe T. Angelo, R. Voltoline, G. Gonçalves, Shin-Ting Wu","doi":"10.24132/jwscg.2021.29.4","DOIUrl":"https://doi.org/10.24132/jwscg.2021.29.4","url":null,"abstract":"As the imaging technology and the understanding of neurological disease improve, a solid understanding of neu-roanatomy has become increasingly relevant. Neuroanatomy teaching includes the practice of cadaveric dissectionand neuroanatomy atlases consisting of images of a brain with its labeled structures. However, the natural inter-individual neuroanatomical variability cannot be taken into account. This work addresses the individual grossneuroanatomy atlas that could enrich medical students’ experiences with various individual variations in anatomi-cal landmarks and their spatial relationships. We propose to deform the CerebrA cortical atlas into the individualanatomical magnetic resonance imaging data to increase students’ opportunity to contact normal neuroanatomicalvariations in the early stages of studies. Besides, we include interactive queries on the labels/names of neu-roanatomical structures from an individual neuroanatomical atlas in a 3D space. An implementation on top ofSimpleITK library and VMTK-Neuro software is presented. We generated a series of surface and internal neu-roanatomy maps from 16 test volumes to attest to the potential of the proposed technique in brain labeling. Forthe age group between 10 to 75, there is evidence that the superficial cortical labeling is accurate with the visualassessment of the degree of concordance between the neuroanatomical and label boundaries.","PeriodicalId":39283,"journal":{"name":"Journal of WSCG","volume":"102 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2021-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78980889","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-01-01DOI: 10.24132/JWSCG.2018.26.1.3
Endre Somogyi
Computational models of biological materials enable researchers to gain insight and make testable predictions of quantitative dynamic responses to stimuli. These models are particularly challenging to develop because biological materials are (1) highly heterogeneous containing both biological cells and complex substances such as extra-cellular medium, (2) undergo structural rearrangement (3) couple biological cells with their environment via chemical and mechanical processes. Existing numerical approaches excel at either describing biological cells or solids and fluids, but have difficulty integrating them into a single simulation approach. We present a novel dynamic non-manifold mesh data structure that naturally represents biological materials with coupled chemical and mechanical processes and structural rearrangement in a unified way.
{"title":"A Dynamic Non-Manifold Mesh Data Structure to Represent Biological Materials.","authors":"Endre Somogyi","doi":"10.24132/JWSCG.2018.26.1.3","DOIUrl":"https://doi.org/10.24132/JWSCG.2018.26.1.3","url":null,"abstract":"<p><p>Computational models of biological materials enable researchers to gain insight and make testable predictions of quantitative dynamic responses to stimuli. These models are particularly challenging to develop because biological materials are (1) highly heterogeneous containing both biological cells and complex substances such as extra-cellular medium, (2) undergo structural rearrangement (3) couple biological cells with their environment via chemical and mechanical processes. Existing numerical approaches excel at either describing biological cells or solids and fluids, but have difficulty integrating them into a single simulation approach. We present a novel dynamic non-manifold mesh data structure that naturally represents biological materials with coupled chemical and mechanical processes and structural rearrangement in a unified way.</p>","PeriodicalId":39283,"journal":{"name":"Journal of WSCG","volume":"26 ","pages":"21-30"},"PeriodicalIF":0.0,"publicationDate":"2018-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6298431/pdf/nihms-999176.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"36803824","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：