The interface to our software is web-based, and uses machine learning to make photo identification of individuals easier. Because we’re using machine learning algorithms, the more examples of photos and individuals of a particular species are in our system, the better the software works. The workflow produces data behind the scenes that trains and improves our models.
The high-level features:
The photo identification workflow for our software starts when new photos are imported. To ensure the best identifications, it starts by finding or detecting each individual in each photo. Rather than using bounding boxes by default, the software will draw an outline around the actual individuals. Or, in machine learning terminology, it will semantically segment the photos.
If our software doesn’t find individuals correctly, or if a researcher wants to make sure the detection is very precise, the detections can easily be added to or refined with our annotation tools. The more work researchers put into ensuring individuals are detected, the better our software can detect individuals in the future.
The next step is to identify the particular individuals from the photos, and either match them up to individuals currently in our library (or database) or add them as new individuals. Our software calculates how different each individual is from any individuals currently in the database. Then, it provides the user/researcher with a ranked listing of those individuals. The researcher can then confirm the individual, which adds the encounter as a cross-reference, or the researcher can add it as a new individual to the database. Again, the more this process is repeated, the more data our machine learning algorithms can learn from, and the better they get.
Our software is meant to help with the photo identification process, but it also helps maintain a library, or database, or individuals and encounter notes. It makes it easy to search and view different examples of each individual, and cross-reference research being done. We also plan to more location based functionality, or “light” GIS-type functionality.
Note: We’re still in development right now. If you’re a researcher/biologist/etc. looking to try out photo identification software, and are interested in working with us to develop the software, please feel free to get in touch with us.