Fun for nerds

[RoGeorge]

Disappointing.
Why? Because I was hoping he'd explain how his random-pixel image-reconstruction thingie worked. That's the fascinating part here.

It is called compressed sensing https://en.wikipedia.org/wiki/Compressed_sensing, and roughly said, it's some ugly math that fits a waveform through a set of samples you have.  Which waveform it fits, out of all the infinite possible waveforms?  The one with the minimum number of spectral components.  How is this done, or how to find that particular waveform of a minimum spectrum?  With mathematicians and computers, something like this:



Why that particular choice, out of all possibilities of "connecting" our samples with a line, why only that waveform with the minimum number of spectral lines works the best?  Why does it restores the original, and not some other image?  Nobody knows, but it works.  A lot of statistics and philosophy and hand-waving.

My interpretation is, because the human mind assigns meaning to correlation/structure/patterns, and correlations are compressible information.  Why a brain is about finding corelations?  Because the brain is a processor meant to predict the future.  One can only predict structured information, by definition randomness can not be predicted.  Why did we evolved a future-prediction machine?  To be more successful at staying alive, to avoid stepping into shitty situations.

Predicting the future is not a goal in itself, it's just the result of natural selection.  The dumbest of our ancestors didn't notice the tiger stripes from the grass stripes, and "dieded" faster than those able to "see it coming and avoid the tiger". 

In conclusion, compressed sensing. 
Because of lions and tigers and bears! 

[coppercone2]

[RoGeorge]

Playing Snake with water
Steve Mould



 

[BrianHG]

Real X-Ray vision....

Disappointing.
Why? Because I was hoping he'd explain how his random-pixel image-reconstruction thingie worked. That's the fascinating part here.

Funny, that's the easiest part to understand.  Though, I did design FPGA based image/video enhancement processors 15 years ago when CPU and GPUS weren't fast/or/programmable enough.

All complete missing data, or voids are filled in with approximations between surrounding data.  Think of all missing pixels where absolutely no reflected X-rays were received at all generates a mask overlay image defining a map of where you need to generate the pixels underneath.  All other pixels are considered true, or accurate.

Note that the math used today is far more sophisticated than what I was doing 20 years ago (We could not imagine using sophisticated floating-point math on images at the time which would involve 2D FFT image interpretation):

     For the a film restoration project I did 2 decades ago, we used red, green, and blue light for the true colors, then IR light at an off angle which illuminated just the scratches and dirt on the film.  With some math, we could then stencil out a negative of those pixels scanned with the IR light and rendered those masked out pixels by their surrounding true accurate colored pixels with nothing more than an average and mix algorithm from the surrounding data.  Even though we didn't do sophisticated interpreted resampling of the stenciled out missing image data, for a lot of the old 8mm and 16mm film movies, the final results were still astounding.

[Analog Kid]

Disappointing.
Why? Because I was hoping he'd explain how his random-pixel image-reconstruction thingie worked. That's the fascinating part here.

It is called compressed sensing https://en.wikipedia.org/wiki/Compressed_sensing, and roughly said, it's some ugly math that fits a waveform through a set of samples you have.  Which waveform it fits, out of all the infinite possible waveforms?  The one with the minimum number of spectral components.  How is this done, or how to find that particular waveform of a minimum spectrum?  With mathematicians and computers, something like this:

Thanks. That's what I'm talkin' 'bout.

I'd like to know more about this, understand the details of how it's implemented. Perhaps in another thread.
Apparently the guy programmed this in his little PIC/Arduino/whatever μC he used, so it can't be all that complicated.

According to his video, the solution is the least number of frequency components that fit the points. Pretty elegant.

[RoGeorge]

My guess is the Arduino was used only to move the sensor and to sample the data, (so to collect those black dots seen in the above animated waveform).  Then, those collected samples are sent from Arduino to a PC, where the best guessed image is calculated and displayed.

Pretty sure the algorithms for compressed sensing reconstruction are already implemented, and available as free software, but I never tried any.  There are a few documenting links in the video description (if you click "watch on YouTube", and on the YouTube page of the video, unroll the text description underneath the video).


Other fun things that can be done with compressed sensing is that you can indirectly "see" around a corner, for example.  By looking at the light diffused by a wall (non-line-of-site vision), so by looking at the scattered light from different angles or from different illuminating patterns, it is possible to reconstruct the image of an illuminated object using just a fotoresistor, and without pointing it towards the object.  The fotoresistor is facing toward another wall. 

(a demo at the minute 9:10 of the second video linked here: https://www.eevblog.com/forum/chat/fun-for-nerds/msg5963141/#msg5963141 , and even funnier, that sensor is not a webcam with millions of pixels, it's just a single photoresitor, so a 1 pixel camera can reconstruct the image of an object it can not see, by staring at a blank piece of paper  ).

[electr_peter]

Other fun things that can be done with compressed sensing is that you can indirectly "see" around a corner, for example.  By looking at the light diffused by a wall (non-line-of-site vision), so by looking at the scattered light from different angles or from different illuminating patterns, it is possible to reconstruct the image of an illuminated object using just a fotoresistor, and without pointing it towards the object.  The fotoresistor is facing toward another wall. 

I do not agree that this is "see around the corner" type of thing, or at least not without multiple caveats which make it much less useful. It is a compressed sensing type approach where object is directly lighted with specific structured light patterns and synchronized sensor detects average reflected light value. Average reflected light can be sensed direct, or bounced from many walls, but it still retains average value. But initial synced light patterns are direct, so it is a slight of hand to call it "see around the corner".
Other approaches use lasers and ultra fast light sensors, which is closer to "see around the corner" idea.

[BrianHG]

Playing Snake with water
Steve Mould



 

OK, that looks like fun...

[coppercone2]

https://imgur.com/gallery/building-magnetic-bridge-with-metal-filings-science-exhibit-iPWy2uP

[coppercone2]

how accurate is this?

[David Hess]

how accurate is this?

The angular diameter is about right, but TON 618 would be 200 times brighter than the sun.

[coppercone2]

OK so you need a good filter

I found a equivalent mechanical of disinformation bots

It's like you put on of these, but on your computer monitor instead of an axle

[Phil1977]

Absolutely stunning to see bridge wire detonations in SloMo:

[RoGeorge]

Replicating The Physics Behind Sandstorm Lightning (ft. 3D Printing Nerd)
Plasma Channel



 

[coppercone2]

https://www.science.org/content/article/peacock-feathers-can-be-lasers

I often wondered if my zirconium coated furnace (reflective coating) acts as a weak laser

[coppercone2]

I wonder if the bee is attempting to become a nuclear power

https://www.msn.com/en-us/science/environmental-science/radioactive-wasp-nests-found-near-nuclear-storage-site-in-south-carolina/ar-AA1JLI7a

[RoGeorge]

The curious origins of imperial and metric units
RobWords



 

[BrianHG]

For those who just need 8 minutes of computational peace...



I realize I or someone else must have posted this years ago, but, it needed refreshing...

[SteveThackery]

Fantastic! I love it.

The sound effects are vaguely reminiscent of my days as a Strowger* maintenance guy.

*For newbies, an electromechanical telephone exchange technology.

[BrianHG]

Fantastic! I love it.

The sound effects are vaguely reminiscent of my days as a Strowger* maintenance guy.

*For newbies, an electromechanical telephone exchange technology.

I know.  To me, it is like a pleasant pace of thought stepped in slow motion.
Sort of like a metronome for the brain.

[BrianHG]

The Math Problem That Defeated Everyone… Until Euler

[RoGeorge]

Why Deep Learning Works Unreasonably Well
Welch Labs



 


But how do AI images/videos actually work? | Guest video by ‪@WelchLabsVideo‬
3Blue1Brown



 

[BrianHG]

 

https://www.youtube.com/shorts/5FLIy6rEF9E

[BrianHG]

Another relay piece of art:

[SteveThackery]

Another relay piece of art:

What a wonderful achievement.