MSL Humor and Other Stuff, and other non-technical chat Options

[jmknapp]

Moving on...

[jmknapp]

The real killer app to drive smart watch adoption:

[Astro0]

For some reason Mars reminds me of Grumpy Cat in this image.



Maybe it's annoyed to have so many spacecraft whizzing around it and messing up its surface.

[brellis]

Lately, my eyes aren't as accurate as they used to be: I just read my UMSF bookmark as Unwanted Spaceflight!

[jmknapp]

Funny bit on the recent Big Bang Theory episode, where Wolowitz as celebrity astronaut throws out the first pitch at a California Angels game using a Mars rover (the crowd was not pleased):

https://www.youtube.com/watch?v=fayK8WGIiyc

[algorithm]

How about this guy, reclining in front of the TV, snacks at hand wearing an 'All-In-One' duvet with a 'Big Slipper'

[jmknapp]

Sol 785 NAVCAM view:



"True color" filter applied:

[RoverDriver]

VOCA provides real-time, multi-channel, shout-down voice communications between project elements. Includes voice instruments and custom voice net set up. Service is high-availability and includes fault detection and correction. This is a picture of portion of the screen of one of our VOCA stations. Quite apropos for Halloween!



Paolo

[jmknapp]

I thought it might be fun to run MSL images through face detection software to see if some "faces" emerge out of the random patterns. Turns out the detections aren't really all that facelike, on the order of this from sol 726:



Still, intriguing to think of what similar automated feature detection might produce, using machine learning, etc.

[Gerald]

You find, what you are looking for, provided the threshold is low enough. That's the same for humans as for machines, automated pareidolia, if you like. That's why "discoveries" usually need a 5-sigma confidence level, meaning "definitive evidence" to be generally accepted as a discovery. But even then, these discoveries can be false positives in rare cases.

[jmknapp]

In general, I'd say that an algorithm that finds what one is looking for is a successful algorithm! Of course, if one is looking for faces (or hippos) on Mars, any positives will necessarily be false, but if the algorithm is designed to detect something that's actually there, say spherules, or meteorites, or Jake Matijevic-type rocks, then ideally the true positives will outnumber the false, with the task of sorting through everything manually reduced accordingly.

[Gerald]

The success depends on appropriate a-priori knowledge. If you know, that there is exactly one "face" in the image, an appropriate maximum-likelihood method - which I presume is implemented by the camera software - can find this "face". Hence the search result should be stated in terms of a conditional probability, taking the a-priori knowledge as the condition, i.e. the probability determined by the automated search needs to be multiplied by the probability, that the a-priori assumptions are correct.

This kind of methods worked in a very simplified way e.g. to find stars in the night shots, using the a-priori knowledge, that there are stars, and that they move apparently in a known direction. But you can also false-positively detect stars that way, if there are none.

In the PDS there are loads of processed images, the criteria of which I'm not always aware. But it looks a bit, as if they would try this on a still rudimentary level for the PDS.
More advanced methods have been applied to determine grain size distributions.

There are loads of applications on Earth.

That's a large field of research. Discussing this seriously is rather technical stuff; a better-suited thread for details would probably be in the context of general image processing, or in a separate thread about data reduction and semantics analysis from images.

[jmknapp]

Well, I did post it in another thread, but a mod moved it here, seeing only humor in it evidently!

As for a priori knowledge, I reminded of a story about Vince Lombardi giving his failing Green Bay Packers a lecture on the basics. He started out with "this is a football."

[Phil Stooke]

Seems to me that testing face detection software on images which certainly do not contain faces makes for a good test of the software. If it detects faces in a Mastcam image of an outcrop on Mars - or a radar image of Titan, or a photo of Mt Everest - then it's not doing a very good job. It's an interesting idea to explore the outer limits of what the software will detect as a face, and maybe tweak settings accordingly.

Phil

[centsworth_II]

... If it detects faces in a Mastcam image of an outcrop on Mars... then it's not doing a very good job....