Google has several geospatial foundation models as shared in their Geospatial Reasoning Interest Form. For anyone curious about the links to each of the models, check under this thread

Google released its Data Science Agent in Colab recently and I tried it out and here’s
the review

Find the notebook link below and the dataset used is from Kaggle Titanic challenge

Enter ML, using previous annotations/digitization & the satellite imagery that served as the basis for Object detection model such as YOLO, DETR etc
The newest of the Object detection models, YOLOv12; an attention centric YOLO framework

Digitization using GIS skills helped in extracting features faster but it is a monotonous, mind numbing task. Clicking on a screen to extract trees or forested areas in an satellite imagery of Nairobi would take several days while being done by several people

Why get the data you ask?

ML in geospatial technologies.
ML have revolutionized various industries & the geospatial industry is one that has gained immensely & will continue to

Think of mapping of various features from EO where satellites are continuously taking images of the earth thus updated data

It features a diverse suite of tasks including summarization, perception, visual reasoning, & navigation, all requiring processing information across extended time periods.

A TikTok on the same:
vm.tiktok.com/ZMkJqRb4e/

I keep forgetting that bluesky exists

This is also on Twitter: HourVideo by Stanford a new benchmark dataset for evaluating AI's ability to understand hour-long videos.