Wildbook blends structured wildlife research with artificial intelligence, citizen science, and computer vision to speed population analysis and develop new insights to help fight extinction. Here is our vision in one minute.
Or if you have a bit more time, check out this video about our work with NVIDIA:
Where's the Wildbook team? If you take a look at our master branch on Github.com and separately in our Image Analysis repo, you'll see commits flying fast and furious. Wildbook 7 is nearing completion, and many of our platforms have already been upgraded to stay current with master. However, our recent support from Microsoft has given us time to rip off old band-aids and remove old hacks. We're re-implementing all existing Wildbooks with a common, new computer vision pipeline that can pivot and behave differently for each species. This has been a huge lift and far more work than we expected, but we're very happy to have the time and funding to finally do it. It's absolutely critical to drive complexity and cost down, and sometimes only experience can show how that should be done. Here is a list of what you can expect in the upcoming release of Wildbook 7.
According to a July 2017 study in the Proceedings of the National Academy of Sciences, a “sixth mass extinction” is underway, a trend signalled by widespread vertebrate losses that “will have negative cascading consequences on ecosystem functioning and services vital to sustaining civilization.” This meta-study is based on multiple, independent analyses and represents a growing awareness in the wildlife research community that more rapid assessment, response, and review are needed to understand and counter this decline.
Unfortunately, wildlife research efforts are frequently underfunded and small scale. The collection and management of wildlife data remains a largely ad hoc and academic exercise focused on moving small data sets (often in Excel and Access) into local, custom population studies for “one-off” analyses without long-term data curation or collaboration across borders and regions. Arriving at a critical mass of data for population analysis can take years (especially for rare or endangered species). Long required observation periods and manual data processing (e.g., matching photos “by eye”) can create multi-year lags between study initialization and scientific results, as well as create conclusions too coarse or slow for effective and optimizable conservation action. This limits the scope, scale, repeatability, continuity, and ROI of the studies as they face the limits of their home-grown tools and IT capabilities.
Wildlife researchers lack a common yet customizable platform for collaboration and often don’t have the technical experience or budget to take advantage of advanced computing tools (e.g., computer vision, artificial intelligence). These tools allows projects to obtain, curate, and analyze “Big Data”, such as the potential of citizen scientists to collect and contribute large volumes of wildlife data through tourism and volunteerism.
Wildbook ® is an open source software framework to support collaborative mark-recapture, molecular ecology, and social ecology studies, especially where citizen science and artificial intelligence can help scale up projects. It is developed by the non-profit Wild Me (PI Jason Holmberg) and research partners at the University of Illinois-Chicago (PI Tanya Berger-Wolf), Rensselaer Polytechnic Institute (PI Charles V. Stewart), and Princeton University (PI Daniel Rubenstein).
Wildbook provides a technical foundation (database, APIs, computer vision, etc.) for wildlife research projects that are:
The biological and statistical communities already support a number of excellent tools, such as Program MARK,GenAlEx, and SOCPROG for use in analyzing wildlife data. Wildbook is a complementary software application that:
Images have become the most abundant, available and cheap source of data. The explosive growth in the use of digital cameras, together with rapid innovations in storage technology and automatic image analysis software, makes this vision possible particularly for large animals with distinctive striped, spotted, wrinkled or notched markings, such as elephants, giraffes and zebras. This large number of collected images must be analyzed automatically to produce a database that records who the animals are, where they are, and when they were photographed. Combining this with geographic, environmental, behavioral and climate data would enable the determination of what the animals are doing, and why they are doing it.
Wildbook evolved out of multi-disciplinary, collaborative research conducted under National Science Foundation support (see ibeis.org). Wildbook employees computer vision and A.I. components to detect features in submitted images and detect and then identify individual animals. Wildbook brings massive-scale computer vision to wildlife research for the first time.
Wildbook integrates the data management software of Wild Me with the computer vision and A.I. research of RPI. Wildbook includes a two-part, multi-species computer vision pipeline to find and identify individual animals in photos collected under real-world conditions, especially with citizen science contribution.
Our detection pipeline is a cascade of deep convolutional neural networks (DCNNs) that applies a fully-connected classifier on extracted features. Three separate networks produce: (1) whole-scene classifications looking for specific species of animals in the photograph, (2) object annotation bounding box localizations, and (3) the viewpoint, quality, and final species classifications for the candidate bounding boxes.
In Wildbook, A.I.-powered detection finds and labels wildlife in photos.
The second major computer vision step is identification, which assigns a name label to each annotation from detection. To do this, SIFT descriptors are first extracted and then compared at keypoint locations. Scores from the query that match the same individual are accumulated to produce a single potential score for each animal. The animals in the database are then ranked by their accumulated scores. A post-processing step spatially verifies the descriptor matches and then re-scores and re-ranks the database individuals.
Example correct identifications. The upper left annotation in each frame is the annotation to be identified. The other frames are the other annotations for the same animal. The bottom left annotation is the primary matching frame. The colored line segments show connections between corresponding features of the same animal.
The results of computer vision are returned to Wildbook’s data management software, which supports rapid curation, export, and analysis. Data can be rapidly viewed in tables, maps, charts, calendars, and as thumbnails. Data can also be searched, filtered, and used in R, Mark, ArcGIS, Google Earth, and other applications.
Wildbook is used in public and private installations.
Wildbook has an R package available at: https://cran.r-project.org/web/packages/RWildbook/index.html
The following support options can help you use Wildbook.
Wildbook® is always free and open source. We are a community of IT professionals and wildlife researchers maintaining and improving a 21st century platform. However, sometimes you may need extra help on a deadline. Our non-profit Wild Me offers professional hosting and customization to fit your project's requirements. This helps us fund our non-profit projects. Contact us if you need help!
Wildbook is a long-term, multi-disciplinary, multi-institution project combining skilled people in computer science, data science, ecology, and software engineering..
Professor Tanya Berger-Wolf provides computer science, data science, and overall project leadership to Wildbook.
Jason Holmberg is the Director of Engineering for Wildbook. Jon Van Oast, Drew Blount, and Colin Kingen are the primary software developers. Jason Parham is our machine learning specialist. Together we bring a wealth of professional software engineering experience to Wildbook.
Professor Charles Stewart and his students provide artificial intelligence and computer vision research and technology to Wildbook.
Hendrik Weideman is completing his PhD on CurvRank, the matching technology that can identify individual animals based on natural edges, such as fin shapes.
Professor Dan Rubenstein provides ecology and biology guidance to Wildbook as well as field testing Kenya.
Dr. Jonathan Crall is one of the original developers of Wildbook's computer vision components under the IBEIS project and specializes in individual animal identification challenges in computer vision.
Dr. Scott Baker of Oregon State University designed the DNA-related components within the software and remains an active adviser on the project.
Ongoing support for Wildbook is funded by Microsoft's A.I. for Earth Program, Pineapple Fund, Amazon Web Services, collaborative co-investment by users, and donations to Wild Me.
Past development work for Wildbook has been supported by:
You can help move Wildbook forward by making a donation! Your donation is tax deductible in the United States.
Can you donate BitCoin or Ethereum?
BitCoin wallet ID: 15KCE1xCGjhcDojjEpMhUA8L6JMdiNGTHy
Ethereum wallet ID: 0x5d41f2e86FeCD1205717B099a8546c5cF6F97e57
The following publications have resulted or been supported by Wildbook-related work:
The following publications have influenced our design and development of Wildbook: