Talk about finding a needle in a hay stack!

*promote

Your hand on his arm
The haystack charm
Around your neck

Strung out and thin
Calling some friend
Trying to cash some check

He's acting dumb
That's what you've come
To expect

Needle in the hay
Needle in the hay
Needle in the hay
Needle in the hay

He's wearing your clothes
Head down to toes
A reaction to you

You say you know what he did
But you idiot kid
You don't have a clue

Sometimes they just
Get caught in the eye
You're pulling him through

Needle in the hay
Needle in the hay
Needle in the hay
Needle in the hay

Now on the bus
Nearly touching
This dirty retreat

Falling out
6th and Powell
A dead sweat in my teeth

Going to walk, walk, walk
Four more blocks
Plus the one in my brain

Down downstairs
To the man
He's going to make it all ok

I can't beat myself
I can't beat myself
And I don't want to talk

I'm taking the cure
So I can be quiet
Whenever I want

So leave me alone
You ought to be proud
That I'm getting good marks

Needle in the hay
Needle in the hay
Needle in the hay
Needle in the hay

Needle in the hay
Needle in the hay
Needle in the hay
Needle in the hay

Tags for this video have been changed from 'kermit, royal tenebaums, suicide, needle in the hay, muppets' to 'kermit, royal tenebaums, suicide, needle in the hay, muppets, sad kermit' - edited by joemawlma

Since the only Elliot Smith tune I have heard previous to this one was "Needle In The Hay", from it's use in "The Royal Tennebaums", I was surprised by this song. I guess I assumed all of his music would sound more like that song than this one.

This is indeed quite interesting, though I honestly don't see any practical applications for it beyond being a contextually superior "Google Images" service.

@haggis: I would be very surprised if the algorithm for discovering points of similarity between images was anything but fussy. It follows intuitively that the greater the resolution of an image, the more bits of information are available from which a more comprehensive pattern -- the sequence of "key points" as described by the engineer duder -- can be abstracted by "filters".

The stitching feature no doubt utilizes something like a Bayesian curve algorithm when comparing patterns of two different images. This is akin to finding a needle in a hay stack, except that the hay stack is made of needles and you only want to find needles possessing particular attributes which match a pattern within some arbitrary threshold. I can only imagine how easily this breaks down with the slightest gap between images.