The gulf to the east of Mexico and to the south of the US, according to Google Maps when VPNing into different countries
1 - US
2 - Mexico
3 - India (or many other third countries)

Or 3 - craters that were already visible in the image from the early part of the conflict, meaning they were possibly created by bombs dropped in a prior conflict.

2 - Detections that were too small (the blue circle, which has a diameter of 39.37 feet was designed to help with this):

Through this workflow, we could filter out false positives, like 1 - Detections that weren’t craters (AI detected regions shown with green outline):

Then, we set up a workflow where for each geotagged box, we had the corresponding Planet satellite image from Nov 16th/17th on the right, and an image from early in the conflict on the left.

We needed to verify these detections. In @qgis.fosstodon.org.ap.brid.gy we drew ~100m boxes around each detection, stitching together detections that were within 100m of each other, so that we were left with 1,185 boxes.

Once trained, we ran the object detection algorithm on @planetlabs.bsky.social satellite images of southern Gaza from November 16th and 17th. We ran 3 iterations of the algorithm, which produced 1,643 detections in all.

Though for our final analysis we were interested in craters roughly ~40ft in diameter or larger, we used these as well as smaller craters in the training data to make the detector more robust. Later, we removed small craters the detector found from our final findings.

Using the platform @picterra.bsky.social, we trained the algorithm with many examples of the types of craters we were looking for. So that the training data most closely resembled the final imagery we would run the algorithm on, we used examples of craters from Gaza itself as training data.

Sometimes 2,000-pound bombs leave craters in the ground where they strike that are approximately 40ft in diameter. Leveraging this fact, we trained an object detection AI algorithm to find craters visible in satellite imagery of south Gaza

And then "indexed" these photos with an open-source algorithm released by OpenAI called CLIP. This gave us the ability to search through the images using simple text queries:

Had the pleasure of speaking to the inaugural Open Source Journalism Conference at USC this weekend about my approach to using AI in journalism

Sea level rise, as envisioned by Severance

from www.reddit.com/user/Unlikel...

A little to the west, east of Temescal Canyon trailhead

Same area zoomed out, with the Will Rogers Polo Field to the left and the Riviera golf course to the right

Drone footage shows the devastating and widespread destruction in LA's Pacific Palisades

1&2 - Area around Galloway St/Bashford St just northeast of Sunset Boulevard

3&4 - Area around Chautauqua Blvd, south west of Will Rogers State Park