An Open Source Framework for Underwater Image Processing
Video and Image Analytics for Marine Environments
In cooperation with NOAA’s Automated Image Analysis Strategic Initiative (AIASI), Kitware has developed VIAME, an open-source system for analysis of underwater video and imagery for fisheries stock assessment. VIAME will enable rapid, low-cost integration of new algorithmic modules, datasets and workflows.
- Open Source basis fosters easy collaboration and transparent development for the widest possible community involvement.
- Multiple Baseline Algorithms allows users to train detector, classification, and size estimation models based on varying amounts of training data.
- Modular Dataflow Architecture enables straightforward integration of a wide variety of third party analytics, libraries and tools.
- Based on KWIVER, the Kitware Video and Image Exploitation and Retrieval toolkit, which includes a pipelined image processing framework, baseline algorithms, and wrappers around algorithms from the broader computer vision community.
- Installers and Quick-Start Guides: Software installers, basic user manuals
- VIAME Developer’s Manual: More comprehensive guide covering building from source and examples
- VIAME Overview: Overview covering VIAME’s goals and design
- NOAA Data Overview: Overview of relevant NOAA collection platforms datasets
- Github Project Page Project development
- VIAME Blog Announcements and News
- Mailing List VIAME Discussion
Due to the delays in rolling out the challenge data, we have extended the submission deadline from May 16 until May 23. The competition results will now be announced on May 30. For information about submissions, please see the previous blog post.
The submission server is now open and accepting submissions. Please see our Submission Guidelines page for information about properly formatting your submission and a step-by-step process of completing a submission.
We're excited to announce that the challenge data is now available for download. The challenge data is structured similarly to the imagery released in the previous training data phase, with folders consisting of imagery for a particular data set. You should submit one...