The Fishial.AI project was created by the Wye Foundation to support the build of Fishial Recognition™️. Just like facial recognition is an artificial intelligence tool that is used to suggest who the person in the image may be, Fishial Recognition™️ will use artificial intelligence to suggest what fish species is in the image. AI programs are built by collecting a wide range of images, training the model with these images, and then testing the model with a fresh set of images. This process is repeated until the model can produce the desired results. An AI program will only “know” as much as it is trained by using training images. This being the case, and wanting to build a Fishial Recognition™️ program robust enough to be used world-wide, the Wye Foundation created Fishial.AI.
The goal of Fishial.AI is to build the world’s largest, open-source, labeled fish image database. The data collected will be used to create an open-source model that uses computer vision to successfully identify fish species within an image. We aim to build a model robust enough to be able to identify fish images from all over the world, and in order to do so we must have a lot of pictures of a lot of fish. Fishial.AI has a portal that will allow users to help us collect all of the images that we need.
The portal feature in Fishial.AI will allow the public to create their own unique username that they will log in to the portal with. Once logged in, users will be able to upload as many fish images as they wish. Users will have the opportunity to tag the fish species in their images, and tag the unique identifying attributes of each fish species. Once the user submits an image, it will get sent to review by the Fishial.AI staff. Once a staff member verifies the data in the image, the image will count towards the users approved images. The status of the image is followed as it moves through the process, taking note the number of uploaded, tagged, submitted and approved images. This allows Fishial.AI to track the progress of all of the users and to create a leader board where users can see where they stand amongst other participants.
All images will use polygonal style bounding boxes for the fish species tag (the whole fish) and for the identifying attributes. We have decided to use this method in hopes that this will reduce background noise for the model and allow for a more efficient and accurate training process. We anticipate that by training the model with images that have been tagged with unique identifying attributes for each fish that we will be able to create a model that will be able to identify fish species within images with high probability.
Fishial.AI seeks to be on the leading forefront of technology to assist in global conservation efforts. Operating as an open-source database, Fishial.AI has the ability to help anyone in need of identified fish images- which can be used to build a myriad of other AI, computer vision and machine learning tools.
