Codahead
We use cookies to optimize your user experience. We also share information about your use of our site with our social media, advertising and analytics partners. By continuing to use our site you agree to use cookies in accordance with our Privacy Policy
Accept
Fishial

AI Fish recognition
platform

Fishial.AI was created by private foundation with mission to
become the world’s largest marine biodiversity-related citizen
science project focused on fish identification.

Fishial
Client
FishAngler
Country
USA
Industry
Education/Science
Year
2019 - present
Fishial

Business needs

  • Platform for the scientific marine and angler communities
  • A tool to attain photos and build AI models for fish recognition
  • First in the world the most complete open-source database of fish
  • Online and mobile access

What we did

Fishial.AI was created by private foundation with mission to become the world’s largest marine
biodiversity-related citizen science project focused on fish identification. With them, we have designed
and implemented a unified platform for the cooperation of various entities to create the world's largest
training database for fish pictures.

We have prepared and created an editor supported by AI/Machine Learning/Computer Vision technology
that allows for automatic shape determination and fish attribute identification. We have developed
AI/Machine Learning models that support the editor and create polygons describing the fish in order to
optimize photo classification time. The platform's development plan includes using the acquired photo
database to build further AI models recognizing any species of fish in the world.

Key features

  • Advanced AI-enhanced photo editor
  • Complex user structure for review process
  • Comprehensive portal summary statistics for admins,
    editors and team owners
  • CSV batch import for one-click upload of several
    hundreds of photos
  • ML-based fish and species recognition
  • Manual and automatic datasets exports
  • Open and public teams for volunteers
  • Intuitive portal and editor tools and wizards for
    beginners with complete documentation
  • Blurring faces

Dedicated for:

  • Anglers
  • Marine Biologists
  • Scientific communities
  • Universities
  • Government institutions
  • Citizen volunteers
image
  • Machine learning

    The Fishial project includes a part that is responsible for detecting fish and drawing its outlines. This algorithm is implemented as part of the ML side project, the following things have been done:

      1.
      Collecting data for training. To train the neural network, it was necessary to prepare a dataset, we decided to find not only fish but also its contours for better subsequent identification of species. We were able to significantly increase the dataset in a short time and without large resources with other techniques.
      2.
      Neural network training. We used the framework to train the neural network and during the tests, the final version was chosen with pre-trained weights on the COCO dataset.
      3.
      The next step was to deploy the model. We chose the Azure ML service, which provides an easy way to deploy the model, a choice of a virtual machine with a graphics card and easy scaling.
      4.
      Additionally, to improve the performance of researchers who will annotate the images, an algorithm was proposed to find the contours of the fish by highlighting only its bounding box. Thus, the user will not spend a lot of time sketching not found fish.
    fishial

    Cloud native

    To enable seamless application development and high performance, we relied on microservices managed by Kubernetes. The architecture was prepared for processing millions of photos, so we used two cloud providers:

      1.
      Google Cloud hosts the portal along with the photos
      2.
      Microsoft Azure hosts the ML server for fish detection

    Benefits

    • Cost and time reduction of photos analysis thanks to AI/Machine Learning/Computer Vision algorithms
    • Easy of use, for non IT person, photos editor online tool
    • The project was recognized by the Con X Tech Prize jury and qualified for the group of 20 finalists of the Con X Tech Prize competition

    Read more Case Studies

    TickPredict

    Cryptocurrency automated trading platform. Brings together multiple trading applications to support proven strategies like market making, momentum scalping and lag.

    Score Split

    Innovative fundraising platform for charities. Join charity-sponsored games for a chance at great cash payouts while supporting a charity of their choice.

    Appetize

    Enterprise cloud-based Point Of Sale platform with hardware. Trusted and used by the world's highest volume businesses.

    Interested in collaborating?