Skip to content

Roadmap

🚧

WIP

This is a work in progress. refresh the page to see any recent changes made

Integrate AI workflows with Argo.¶

This allows one to run a given workflow from start to finish with a single click

  • ✅ Video processing DONE!
  • 🚧 Image processing
  • 🚧 Audio processing

Below is a video of a workflow in action that runs a detection, classification and tracking model pipeline. To handle huge datasets and models, this uses using Kubernetes to scale our compute. This is currently done by 4 GPUs at MBARI.

roadmap_video_argo.gif

More

Scaling up with Kubernetes offsite¶

GPUs are available beyond MBARI in cloud providers and partner institutions. We are exploring this option to expand our compute capabilities.

Object Segmentation¶

Adding segmentation algorithms for projects that need to calculate the exact area of an object. This is particularly useful for the UAV project kelp coverage studies.

Kelp coverage with Facebook Segment Anything Model (SAM)¶

The SAM model is a state-of-the-art object segmentation model that can be used to find an segment objects with a text prompt.

Click on the image to see a larger version.

kelp_sam_overlay_sm.png

Better Algorithms¶

Labeling is always a challenge and often the bottleneck for machine learning models. Self-supervised learning algorithms can learn from unlabeled data and improve performance. It is the core of many state-of-the-art deep learning models, including DINO which has been used extensively in most of our projects as a backbone. It is a "no labels" approach, meaning it does not require labeled data.

You can check out this MBARI intern paper to see how we’ve tested this on AUV datasets.

last updated: 2026-05-03