Saul Greenberg
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Saul Greenberg has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
Theresa Godin
Freshwater Fisheries Society of BC Research Evaluation & Development Section University of British Columbia Vancouver BC Canada
2 shared publications- Anthony Tang
Singapore Management University, Singapore, Singapore
1 shared publication - Bin Zhu
Microsoft
1 shared publication - Graham William Taylor
University of Guelph
1 shared publication Jesse Whittington
Parks Canada, Banff National Park Banff AB Canada
1 shared publication- Jiawei Gu
Baidu Institute of Deep Learning
1 shared publication
Latest Publications
- Top Cited PapersOpen Access
- 7 March 2020
- Published by Wiley in Ecology and Evolution
- Vol. 10 (7) , 3503-3517
Three critical factors affecting automated image species recognition performance for camera traps
Ecological camera traps are increasingly used by wildlife biologists to unobtrusively monitor an ecosystems animal population. However, manual inspection of the images produced is expensive, laborious, and time‐consuming. The success of deep learning systems using camera trap images has been previously explored in preliminary stages. These studies, however, are lacking in their practicality. They are primarily focused on extremely large datasets, often millions of images, and there is little to no focus on performance when tasked with species identification in new locations not seen during training. Our goal was to test the capabilities of deep learning systems trained on camera trap images using modestly sized training data, compare performance when considering unseen background locations, and quantify the gradient of lower bound performance to provide a guideline of data requirements in correspondence to performance expectations. We use a dataset provided by Parks Canada containing 47,279 images collected from 36 unique geographic locations across multiple environments. Images represent 55 animal species and human activity with high‐class imbalance. We trained, tested, and compared the capabilities of six deep learning computer vision networks using transfer learning and image augmentation: DenseNet201, Inception‐ResNet‐V3, InceptionV3, NASNetMobile, MobileNetV2, and Xception. We compare overall performance on “trained” locations where DenseNet201 performed best with 95.6% top‐1 accuracy showing promise for deep learning methods for smaller scale research efforts. Using trained locations, classifications with <500 images had low and highly variable recall of 0.750 ± 0.329, while classifications with over 1,000 images had a high and stable recall of 0.971 ± 0.0137. Models tasked with classifying species from untrained locations were less accurate, with DenseNet201 performing best with 68.7% top‐1 accuracy. Finally, we provide an open repository where ecologists can insert their image data to train and test custom species detection models for their desired ecological domain.
- 7 March 2020
- Published by Wiley in Ecology and Evolution
- Vol. 10 (7) , 3503-3517
Top Cited PapersOpen AccessThree critical factors affecting automated image species recognition performance for camera traps
Ecological camera traps are increasingly used by wildlife biologists to unobtrusively monitor an ecosystems animal population. However, manual inspection of the images produced is expensive, laborious, and time‐consuming. The success of deep learning systems using camera trap images has been previously explored in preliminary stages. These studies, however, are lacking in their practicality. They are primarily focused on extremely large datasets, often millions of images, and there is little to no focus on performance when tasked with species identification in new locations not seen during training. Our goal was to test the capabilities of deep learning systems trained on camera trap images using modestly sized training data, compare performance when considering unseen background locations, and quantify the gradient of lower bound performance to provide a guideline of data requirements in correspondence to performance expectations. We use a dataset provided by Parks Canada containing 47,279 images collected from 36 unique geographic locations across multiple environments. Images represent 55 animal species and human activity with high‐class imbalance. We trained, tested, and compared the capabilities of six deep learning computer vision networks using transfer learning and image augmentation: DenseNet201, Inception‐ResNet‐V3, InceptionV3, NASNetMobile, MobileNetV2, and Xception. We compare overall performance on “trained” locations where DenseNet201 performed best with 95.6% top‐1 accuracy showing promise for deep learning methods for smaller scale research efforts. Using trained locations, classifications with <500 images had low and highly variable recall of 0.750 ± 0.329, while classifications with over 1,000 images had a high and stable recall of 0.971 ± 0.0137. Models tasked with classifying species from untrained locations were less accurate, with DenseNet201 performing best with 68.7% top‐1 accuracy. Finally, we provide an open repository where ecologists can insert their image data to train and test custom species detection models for their desired ecological domain.
- Open Access
- 2 December 2019
- Published by Wiley in Ecology and Evolution
- Vol. 9 (24) , 13706-13730
Design patterns for wildlife‐related camera trap image analysis
This paper describes and explains design patterns for software that supports how analysts can efficiently inspect and classify camera trap images for wildlife‐related ecological attributes. Broadly speaking, a design pattern identifies a commonly occurring problem and a general reusable design approach to solve that problem. A developer can then use that design approach to create a specific software solution appropriate to the particular situation under consideration. In particular, design patterns for camera trap image analysis by wildlife biologists address solutions to commonly occurring problems they face while inspecting a large number of images and entering ecological data describing image attributes. We developed design patterns for image classification based on our understanding of biologists' needs that we acquired over 8 years during development and application of the freely available Timelapse image analysis system. For each design pattern presented, we describe the problem, a design approach that solves that problem, and a concrete example of how Timelapse addresses the design pattern. Our design patterns offer both general and specific solutions related to: maintaining data consistency, efficiencies in image inspection, methods for navigating between images, efficiencies in data entry including highly repetitious data entry, and sorting and filtering image into sequences, episodes, and subsets. These design patterns can inform the design of other camera trap systems and can help biologists assess how competing software products address their project‐specific needs along with determining an efficient workflow.
- 2 December 2019
- Published by Wiley in Ecology and Evolution
- Vol. 9 (24) , 13706-13730
Open AccessDesign patterns for wildlife‐related camera trap image analysis
This paper describes and explains design patterns for software that supports how analysts can efficiently inspect and classify camera trap images for wildlife‐related ecological attributes. Broadly speaking, a design pattern identifies a commonly occurring problem and a general reusable design approach to solve that problem. A developer can then use that design approach to create a specific software solution appropriate to the particular situation under consideration. In particular, design patterns for camera trap image analysis by wildlife biologists address solutions to commonly occurring problems they face while inspecting a large number of images and entering ecological data describing image attributes. We developed design patterns for image classification based on our understanding of biologists' needs that we acquired over 8 years during development and application of the freely available Timelapse image analysis system. For each design pattern presented, we describe the problem, a design approach that solves that problem, and a concrete example of how Timelapse addresses the design pattern. Our design patterns offer both general and specific solutions related to: maintaining data consistency, efficiencies in image inspection, methods for navigating between images, efficiencies in data entry including highly repetitious data entry, and sorting and filtering image into sequences, episodes, and subsets. These design patterns can inform the design of other camera trap systems and can help biologists assess how competing software products address their project‐specific needs along with determining an efficient workflow.
- Open Access
- 30 August 2015
- Published by Springer Nature
- p. 436-454
CipherCard: A Token-Based Approach Against Camera-Based Shoulder Surfing Attacks on Common Touchscreen Devices
- 30 August 2015
- Published by Springer Nature
- p. 436-454
Open AccessCipherCard: A Token-Based Approach Against Camera-Based Shoulder Surfing Attacks on Common Touchscreen Devices
- 1 June 2015
- Published by Oxford University Press (OUP) in Fisheries
- Vol. 40 (6) , 276-287
A Tool Supporting the Extraction of Angling Effort Data from Remote Camera Images
Estimating angling effort on more than a few lakes can be prohibitively expensive using creel surveys and often requires finer-scale angler distribution data than aerial surveys can provide. An alternate method uses remote cameras to capture images of lakes at hourly intervals over long time periods. Technicians then visually analyze the thousands of generated images for features of interest (e.g., angler counts and environmental conditions) and use those data to estimate angling effort. The problem is that the visual analysis step is time-consuming, expensive, and difficult to validate. Consequently, we elicited the strategies and best practices technicians used when analyzing images and identified bottlenecks. We then designed software, called Timelapse to better support image analysis. In use for several years, Timelapse has proven a cost-effective method of estimating angling effort in British Columbia's small lakes fisheries; it significantly eases a technician's workflow and doubles the number of images one can process per hour. La estimación del esfuerzo de pesca con anzuelo que se realiza en varios lagos mediante muestreos en puerto, puede llegar a tener costos prohibitivos y suele requerir una información más fina sobre la distribución del esfuerzo que la que proveen los muestreos aéreos. Un método alternativo se basa en la captura de imágenes de los lagos usando cámaras remotas, tomadas cada hora durante largos periodos. Posteriormente, los cientos de imágenes generadas son analizadas visualmente por los técnicos con el fin de detectar características de interés (v.g. conteo de pescadores y condiciones ambientales) y esta información se usa para estimar el esfuerzo de pesca. El problema es que el análisis visual consume mucho tiempo, es caro y difícil de validar. En consecuencia, en este trabajo se elucidan las estrategias y mejores prácticas que el personal técnico utiliza al analizar las imágenes e identificar cuellos de botella. Posteriormente se diseña un programa llamado Timelapse, como apoyo para el análisis de imágenes. Habiendo sido utilizado por varios años, Timelapse ha mostrado ser un método efectivo en cuanto a costos para estimar el esfuerzo en pesquerías de pequeños lagos en la Columbia Británica; alivia de forma importante el flujo de trabajo del personal técnico y duplica el número de imágenes que pueden procesarse en una hora.
- 1 June 2015
- Published by Oxford University Press (OUP) in Fisheries
- Vol. 40 (6) , 276-287
A Tool Supporting the Extraction of Angling Effort Data from Remote Camera Images
Estimating angling effort on more than a few lakes can be prohibitively expensive using creel surveys and often requires finer-scale angler distribution data than aerial surveys can provide. An alternate method uses remote cameras to capture images of lakes at hourly intervals over long time periods. Technicians then visually analyze the thousands of generated images for features of interest (e.g., angler counts and environmental conditions) and use those data to estimate angling effort. The problem is that the visual analysis step is time-consuming, expensive, and difficult to validate. Consequently, we elicited the strategies and best practices technicians used when analyzing images and identified bottlenecks. We then designed software, called Timelapse to better support image analysis. In use for several years, Timelapse has proven a cost-effective method of estimating angling effort in British Columbia's small lakes fisheries; it significantly eases a technician's workflow and doubles the number of images one can process per hour. La estimación del esfuerzo de pesca con anzuelo que se realiza en varios lagos mediante muestreos en puerto, puede llegar a tener costos prohibitivos y suele requerir una información más fina sobre la distribución del esfuerzo que la que proveen los muestreos aéreos. Un método alternativo se basa en la captura de imágenes de los lagos usando cámaras remotas, tomadas cada hora durante largos periodos. Posteriormente, los cientos de imágenes generadas son analizadas visualmente por los técnicos con el fin de detectar características de interés (v.g. conteo de pescadores y condiciones ambientales) y esta información se usa para estimar el esfuerzo de pesca. El problema es que el análisis visual consume mucho tiempo, es caro y difícil de validar. En consecuencia, en este trabajo se elucidan las estrategias y mejores prácticas que el personal técnico utiliza al analizar las imágenes e identificar cuellos de botella. Posteriormente se diseña un programa llamado Timelapse, como apoyo para el análisis de imágenes. Habiendo sido utilizado por varios años, Timelapse ha mostrado ser un método efectivo en cuanto a costos para estimar el esfuerzo en pesquerías de pequeños lagos en la Columbia Británica; alivia de forma importante el flujo de trabajo del personal técnico y duplica el número de imágenes que pueden procesarse en una hora.
- Open Access
- 18 April 2015
- Published by Association for Computing Machinery (ACM)
- p. 2479-2480
Sketching User Experiences
- 18 April 2015
- Published by Association for Computing Machinery (ACM)
- p. 2479-2480
Open AccessSketching User Experiences
- 1 December 2004
- Published by Springer Nature in Computer Supported Cooperative Work (CSCW)
- Vol. 13 (5-6) , 501-537
Roles of Orientation in Tabletop Collaboration: Comprehension, Coordination and Communication
- 1 December 2004
- Published by Springer Nature in Computer Supported Cooperative Work (CSCW)
- Vol. 13 (5-6) , 501-537
Roles of Orientation in Tabletop Collaboration: Comprehension, Coordination and Communication
- 29 January 1993
- Published by Cambridge University Press (CUP)
- p. 123-140
Organizing activities through workspaces
In every trade a specific way of organizing tools and objects for the craftsman has been established. Every workshop is equipped with appropriate tools and organized with respect to the specific working situation. In this way strategies for the solution of typical problems are at hand for the workers. — Hoffman and Voider, 1986 This book opened by advocating the common metaphor of tool use for thinking about command-based systems, where command lines are the tools people employ to manipulate the materials in their computer environment. The four preceding chapters pursued the notion that recently used lines, like tools, should be available for reuse. But reuse is not the only strategy for supporting user activities. It is evident that people impose some organization on their computer tools and materials, just as craftsmen do with their physical counterparts. Real workshops support these organizations through toolboxes for arranging and locating tools, workbenches for performing specific tasks, shelving and drawers for keeping relevant tools and materials readily available, and so on. Computing environments, on the other hand, do little to promote personal organization. A command-based interface is comparable to an unhelpful clerk who waits for you to name the tool you want, retrieves the tool (if available) from a separate room, and demands that you return it immediately after use. At the other extreme, arranging facilities into fixed taxonomic menus is reminiscent of a totalitarian chaining of tools to a single location.
- 29 January 1993
- Published by Cambridge University Press (CUP)
- p. 123-140
Organizing activities through workspaces
In every trade a specific way of organizing tools and objects for the craftsman has been established. Every workshop is equipped with appropriate tools and organized with respect to the specific working situation. In this way strategies for the solution of typical problems are at hand for the workers. — Hoffman and Voider, 1986 This book opened by advocating the common metaphor of tool use for thinking about command-based systems, where command lines are the tools people employ to manipulate the materials in their computer environment. The four preceding chapters pursued the notion that recently used lines, like tools, should be available for reuse. But reuse is not the only strategy for supporting user activities. It is evident that people impose some organization on their computer tools and materials, just as craftsmen do with their physical counterparts. Real workshops support these organizations through toolboxes for arranging and locating tools, workbenches for performing specific tasks, shelving and drawers for keeping relevant tools and materials readily available, and so on. Computing environments, on the other hand, do little to promote personal organization. A command-based interface is comparable to an unhelpful clerk who waits for you to name the tool you want, retrieves the tool (if available) from a separate room, and demands that you return it immediately after use. At the other extreme, arranging facilities into fixed taxonomic menus is reminiscent of a totalitarian chaining of tools to a single location.
- 29 January 1993
- Published by Cambridge University Press (CUP)
- p. 1-10
Introduction
The Computer User as Toolsmith - January 1993
- 29 January 1993
- Published by Cambridge University Press (CUP)
- p. 1-10
Introduction
The Computer User as Toolsmith - January 1993
- Open Access
- 1 March 1990
- Published by Association for Computing Machinery (ACM) in ACM SIGOIS Bulletin
- Vol. 11 (2-3) , 227-237
Sharing views and interactions with single-user applications
Although work is frequently collaborative, most computer-based activities revolve around software packages designed to be used by one person at a time. To get around this, people working together often talk and gesture around a computer screen, perhaps taking turns interacting with the running “single-user” application by passing the keyboard around. However, it is technically possible to share these unaltered applications—even though they were originally designed for a single user only—across physically different workstations through special view-sharing software . Each person sees the same image of the running application on their own screen, and has an opportunity to interact with it by taking turns. This paper discusses the various roles and responsibilities of the view-sharing software that must be considered during its design and evaluation: view management, floor control, conference registration by participants, and handling of meta-level communications. A brief survey of existing shared view systems is provided for background.
- 1 March 1990
- Published by Association for Computing Machinery (ACM) in ACM SIGOIS Bulletin
- Vol. 11 (2-3) , 227-237
Open AccessSharing views and interactions with single-user applications
Although work is frequently collaborative, most computer-based activities revolve around software packages designed to be used by one person at a time. To get around this, people working together often talk and gesture around a computer screen, perhaps taking turns interacting with the running “single-user” application by passing the keyboard around. However, it is technically possible to share these unaltered applications—even though they were originally designed for a single user only—across physically different workstations through special view-sharing software . Each person sees the same image of the running application on their own screen, and has an opportunity to interact with it by taking turns. This paper discusses the various roles and responsibilities of the view-sharing software that must be considered during its design and evaluation: view management, floor control, conference registration by participants, and handling of meta-level communications. A brief survey of existing shared view systems is provided for background.