Advanced search

Message boards : Number Crunching : Matched Observations ..

Author Message
Profile Gibson Praise
Send message
Joined: 5 Aug 16
Posts: 56
Combined Credit: 50,846,655
DNA@Home: 0
SubsetSum@Home: 73,256
Wildlife@Home: 50,773,399
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 769
Images Observed: 25,032

Message 7173 - Posted: 29 Jun 2017, 3:12:25 UTC
Last modified: 29 Jun 2017, 3:18:37 UTC

So I've noticed on the Top Image Reviewers page that there is now a column entitled "Matched Observations". As near as I can tell this is a stat associated only with the UAV imagery ( since tom7up has no matched observations at all and he hasn't done UAV ).

Are the matches with each other or with software (or both)?

2nd Q: If it is only UAV will it be expanded to the other two projects?

Send message
Joined: 23 Nov 14
Posts: 15
Combined Credit: 5,746,525
DNA@Home: 101,983
SubsetSum@Home: 100,103
Wildlife@Home: 5,544,440
Wildlife@Home Watched: 8,771,275s
Wildlife@Home Events: 2,515
Climate Tweets: 3,044
Images Observed: 43,028

Message 7174 - Posted: 29 Jun 2017, 19:28:08 UTC

Seems so.
I've reviewed 2 UAVs and now I have 2 matched observations.

Marshall Mattingly
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 3 Feb 16
Posts: 27
Combined Credit: 74,756,298
DNA@Home: 0
SubsetSum@Home: 0
Wildlife@Home: 74,756,298
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 1,335

Message 7197 - Posted: 11 Jul 2017, 21:27:08 UTC - in response to Message 7173.

Currently, the matched observations are against other citizen scientist observations. I wonder if I can get a tool-tip in there to define the terms a little bit.

Basically, it checks all of your observations in an image against all observations with other citizen scientists in the image. If all four corners in your observation and another observation are within 10-pixels of each other, we consider that a match -- there's a paper available on how we determined that to be the optimal matching algorithm for this set, compared to a few other choices and parameters.

We're currently doing analysis on expert observations (made by people Susan, et al have trained specifically), raw citizen scientist observations, and matched citizen science observations. We just submitted another paper on that, but the basic findings is that raw citizen science observations are to variable on the same object to make good machine learning candidates, but when matched together, they're nearly as good as the experts.

In the future, we're going to be also matching up against expert observations and observations that our machine learning has detected. I'd like to get some nice web pages together so we can make stats, draw boxes over images, etc publically accessible to get some nice visualization.

Post to thread

Message boards : Number Crunching : Matched Observations ..