Apply dog vision to photo of Robert S. McNamara | Hacker News Reader

Apply dog vision to photo of Robert S. McNamara

wolframalpha.com

93 pointsvinnyglennon80 comments

This sums up the new Wolfram Language in one command - ways of combining data and algorithms in novel/interesting, but practically useless ways.

Well, it's Alpha, not the Wolfram Language.

But I don't think it takes much imagination to see how things like http://www.wolfram.com/language/gallery/visualize-celebrity-... or http://www.wolfram.com/language/gallery/plan-a-city-tour/ or http://www.wolfram.com/language/gallery/track-a-gps-equipped... or http://www.wolfram.com/language/gallery/email-freeze-alerts/ or http://www.wolfram.com/language/gallery/classify-images-as-d... could be the basis of something commercially valuable or publicly useful, especially given how easy they are to create. (edit: and those examples are not even designed for commercial appeal, just concision and cool factor).

And it still has glaring omissions. One thing it still can't do is group theory. I'm not a mathematician, but I occasionally need some group theory stuff at work. This is stuff that is either tabulated or can be calculated easily by a computer. Like "show me a matrix representation of SU(3)" or "decompositions of SU(5)" or "irreps of SU(10)" or something. Group theorists have these little equations with circled operators, like 1+8=3 (imagine numbers in bold and a circle around the plus. I just made up the numbers, but they're usually as wierd), that I can never wrap my head around no matter how often I've tried. But they are kind of important for particle physics, and I hope that wolfram alpha will eventually incorporate that stuff.

Actual particle physics is also lacking, everything besides simple queries like "mass of the top quark" is missing (and the mass it gives is not up-to-date). I would like it to have "decay modes of neutralino-1 at m0=1000, m12=1000, ..." or "higgs mass=126, particle-such-and-such-mass=1000, solve for parameter-x" and it would show me valid values for x resulting in these masses. This is all easily implementable, but I guess the market for it is too small (otoh, I don't know what the market for "dog vision" is...(

I tried replacing dog by cat, and it doesn't work. Why?

Also, for the inverse gamma function of complex numbers: gamma^-1(1+10i) works. gamma^-1(1-10i) confuses its parser (doesn't know how to interpret input, even though the previous and next one work). gamma^-1(1-10.0i) works again. Why?

Cat worked for me.

I tried again, and still doesn't work, it says: Using closest Wolfram|Alpha interpretation: photo of Robert S. McNamara (so it ignores the "cat" part. the fact that the image it shows has a strange yellow color is a coincidence)

No it really doesn't. Think about the underlying abstraction. The point is that somewhere, the engine knew how to define a transformation "dog vision" on a set of objects, namely photos.

Taking it further, it's possible that "dog" was a transformation on "vision" and other visions are available.

As a math major who used Wolfram alpha for the last 3 years of my undergrad some odd years ago, I have to say I had an overwhelmingly positive impression of the product.

Fill in anyone you want:

https://www.wolframcloud.com/objects/032f062e-5a63-4e3a-aff4...

Corresponding code:

  DogVision[img : (_Image | _Graphics | _Graphics3D)] := 
    Module[{rgbimg = ColorConvert[img, "RGB"], data}, 
     data = Transpose[ImageData[rgbimg], {3, 2, 1}];
     GaussianFilter[
      Image[Transpose[{(data[[1]] + data[[2]])/
          2, (data[[1]] + data[[2]])/2, data[[3]]}, {3, 2, 1}]], 5]];

  CloudDeploy[
   FormFunction[{"person" -> "Person"}, 
    DogVision[EntityValue[#person, "Image"]] &, "JPEG"], 
    Permissions -> "Public"]

Every time I play around with Wolfram Alpha, I get frustrated trying to figure out how I get past the natural language input interpretation and write code like you specify here. Is this a restricted feature for pro subscribers? Why do they bother letting me see the "input interpretation" (and provide it as copyable plaintext) if they're not going to let me actually use it?

It sounds like you want to use the Wolfram Language rather than Wolfram Alpha.

I believe what I want is the ability to use Wolfram Alpha by writing my queries in the Wolfram Language. How does one do that?

That's exactly what the Wolfram language does. You cloud deploy your code to use it from the web if you want it to be a service, like REST

Let me be super explicit. Given: I have a piece of Wolfram Language code. Where do I enter it to get it evaluated? Assume I know nothing.

I can't tell if the answer is: A) there's a text box on the web site you are apparently overlooking you moron, B) you buy/download the software, save your code in a text file, and run it using software on your local machine, C) you sign up for Wolfram Alpha Pro Deluxe which then makes the query input box on the Wolfram Alpha page start accepting Wolfram Language, or D) none of the above: [fill in the blank]

You will soon be able to use any of the cloud products mentioned on http://www.wolframcloud.com/.

There will also be Wolfram Desktop, the local IDE for the Wolfram Language that comes with our online products, as well as Mathematica 10, a non-cloud-enabled instantiation of the language for the academic market.

If you want notification as soon as we go public, sign up on http://www.wolfram.com/programming-cloud/.

Disclosure: I deployed this code from my own copy of a prerelease version of Wolfram Desktop, which I could do because I'm an employee.

P.S. Your idea of being able to write Wolfram Language code straight into Alpha and have it execute is an interesting one. You can do some basic stuff like http://www.wolframalpha.com/input/?i=Table[i^2%2C+{i%2C+1%2C...] already, but I agree we should allow you to execute anything (safe) using Alpha.

> P.S. Your idea of being able to write Wolfram Language code straight into Alpha and have it execute is an interesting one.

It's really shocking to me that this idea might be new. To anyone. As an end user, the "input interpretation" box sends a very clear (but apparently wrong) message: "you entered a for-dummies query, which would have looked like [this] in advanced mode." I am pretty sure that every single time I've looked at a Wolfram Alpha result page, I've spent a significant amount of time searching in vain for the way to uncover that advanced mode. It is so absurdly obvious that it must be hidden in there somewhere, because otherwise, what purpose could showing me the raw Wolfram Language text possibly serve? It can only ever be run in an environment that I don't have access to! Alas, there apparently has been no purpose all this time.

I'm working on an open source knowledge engine thing roughly similar to WolframAlpha.

I'm pretty interested in the types of things you might like to do in your "advanced mode".

I've begun work on a SymPy part of it for the math parts, but one of the first things I realised was that it might make sense to expose the "knowledge base" itself to python.

> I'm pretty interested in the types of things you might like to do in your "advanced mode".

"Advanced mode" in this context simply means "I want to write an expression in a formal language because it lets me remove the difficult-to-control natural language parsing." It's actually a simpler mode, because it does less, but it requires the user to use a programming language, which is generally considered "advanced" on this type of user interface.

The answer you're looking for and that everyone is avoiding is Mathematica. I mean, they're working on (or have possibly released by now) other platforms where the so-called Wolfram Language can be used, but the simple answer to "where can I write this natural language stuff or Wolfram Language code" is the application Mathematica.

If I have Mathematica, does that allow me to execute code against the giant corpus of Wolfram Alpha knowledge? That is, will I be able to evaluate 'EdgeDetect[The Edge | image]' and have it use a picture of The Edge as input? Will I be able to just reach out into the cloud and use a weather dataset as input? I was under the impression that the answer is no.

They have a real communication problem, because I inevitably find myself trying to cobble together a notion of what their products actually are. Here's my best guess, rather hard-earned by going in frustrating circles in what passes for documentation on their site:

1) Mathematica is the engine which runs Wolfram Language code. 2) Wolfram Alpha is a combination of (a) Mathematica, (b) a huge knowledge base, and (c) a natural language processor which turns user-provided queries into Wolfram Language expressions. These three things are bundled up with a web front-end and allow you to do really cool stuff. 3) It is not currently possible for me to write my own Wolfram Language expressions to run against the giant knowledge base.

>If I have Mathematica, does that allow me to execute code against the giant corpus of Wolfram Alpha knowledge? Yes. you can use results of natural inputs as a computable data. For example, type ==dog picture, this will give you exactly the same result as Wolfram|Alpha. then you can click on "+" sign in the top right corner of the pod and select "subpod content". which will give something like WolframAlpha["dog picture", {{"Image:SpeciesData", 1}, "Content"}] and it is an image you can do processing on.

(I'm very much an amateur.)

You should be able to evaluate something very similar. Typing EdgeDetect[CTRL = The Edge image] (where "CTRL =" indicates "execute the key combination CTRL and Equals key") should give you the result in anything which runs the Wolfram Language. I believe that will include Mathematica and also the Wolfram Desktop. WolframAlpha is an instantiation of the Wolfram Language with a huge knowledge base and a parser - it used to be Mathematica+knowledgebase+parser, but now it's explicitly got the cloud capabilities of the Wolfram Language which I am under the impression that Mathematica will not have.

Note that you can access Alpha from Mathematica (by dint of the command WolframAlpha, or by typing two Equals signs before what you type), but you can only enter the crudest Wolfram Language commands into Alpha and have them executed. However, using the Mathematica tie-in as above, it is possible to write your own Mathematica expressions (and soon Wolfram Language expressions, when it's released properly) which make use of the Alpha knowledgebase.

I don't know much about this, but I've seen a talk with Stephen Wolfram before they released this - I think, based on how he described it, that D is the correct answer. I believe it's more similar to writing an application than just entering a query like you would on their homepage. So I guess in a sense C is correct, but I don't think that's quite how it works.

You buy Mathematica and run it there.

Almost anyone. I can't find the terms to get it to show me how a dog sees a postal carrier. Various terms either failed, or applied the vision to a film poster instead.

Postal worker is probably not an entity of type Person, I would say it's more of a profession.

This stuff is what really irritates me about Wolfram Alpha. Almost everything they do beyond basic unit conversion comes across as a nifty parlor trick instead of a real capability.

Like you see this example, and think "Oh, it can apply different types of animal vision to different types of images." so you try this:

http://www.wolframalpha.com/input/?i=apply+cat+vision+to+pho...

And you get:

> Input interpretation: Canis Minor | constellation region

With a diagram of the constellation Canis Minor...which I guess is SORT of a small dog.

But this kind of stuff is rife throughout Wolfram Alpha, and makes it much much less useful than Google. For example, here's a question I was curious about recently: http://www.wolframalpha.com/input/?i=world+energy+use+per+ye...

Wolfram just totally ignores the second part of the query, versus Google[0], which gives this very on-topic result for the third result: http://www.sandia.gov/~jytsao/Solar%20FAQs.pdf‎

[0] https://encrypted.google.com/search?hl=en&q=world%20energy%2...

The Edges of The Edge:

http://www.wolframalpha.com/share/clip?f=d41d8cd98f00b204e98...

The Boss Embossed:

http://www.wolframalpha.com/input/?i=apply+embossing+effect+...

I was sure it would be http://www.wolframalpha.com/input/?i=apply+embossing+effect+...

Haha awesome. "Blur Blur" should work, I've filed a bug :-)

Had no idea that dogs' visual acuity was so poor. My dog seems to be able to spot squirrels at 100 yards...

From what I understand, both dogs and cats have poor "static" vision but can spot minute changes very, very well. A squirrel has a very specific gait and movement style which they can spot very well. If your dog saw a photo of a squirrel from 100yrds away, I doubt it would heed much attention to it.

My dog can also spot food on the cement sidewalk from far away, too far to be just the smell.

Recent research[1] has found a fovea like structure on the canine retina. Fovea is a small patch of the retina with increased density of visual receptors. The finding suggests that the dogs has better visual acuity than previously suspected.

Quite likely this visualization were programed before this new development.

[1]: http://www.sciencedaily.com/releases/2014/03/140305191513.ht...

Dogs are hunters, not foragers. They spot movement very well indeed, and a squirrel's hop -- pause -- hop -- pause　movement style is very easy to recognise.

Perhaps smell is in there too, and perceiving movement better?

Spotting movement is a different receptor.

Maybe smell?

Is Smell directional in dogs?

I think the nostrils are too close together for 3d smell detection beyond very short odor wavelengths.

Perhaps if the dog is also aware of the direction of the wind during the time of the smell.

With whiskers, it might actually work in that way...

Yeah, I honestly suspect they might be able to do that. Human hunters often pay attention to wind direction when stalking prey so that they don't give away their position. This would be fairly similar to that.

yes, it's one of the reasons they articulate and constantly wet their nostrils.

http://chemse.oxfordjournals.org/content/30/4/291.full

It's not so poor. https://news.ycombinator.com/item?id=7441530

Replacing "dog" with "cat" does not work... though the resulting interpretation ("photo of Robert S. McNamara") is oddly yellowed.

I tried exactly the same thing. I always find WolframAlpha frustrating for anything except arithmetic. Example: search "sat scores" and you get some info about SAT scores, including a histogram by gender. Search "sat scores male" and the query is not understood, and you get the page for SAT scores. Search "sat scores of males" and the query is not understood, and you get the page for "sat male", i.e. a page describing an infant born tomorrow.

It's reminiscent of the old text adventures, where the easy part was figuring out what to do, and the hard part was figuring out the right syntax to match the coder's expectations.

No wonder cats and dogs don't get along very well...

http://www.wolframalpha.com/input/?i=apply+dog+vision+to+pho...

Inverted color http://www.wolframalpha.com/input/?i=invert+photo+of+Robert+...

Upsidedown http://www.wolframalpha.com/input/?i=flip+photo+of+Robert+S....

Black and white http://www.wolframalpha.com/input/?i=grayscale+photo+of+Robe...

Fog of war?

All kidding aside it's interesting to know that the language can perform these kinds of transformations but I'm not quite sure what my use case would be yet.

Whenever it gets radically easier to do higher-level things, it usually takes a while for people to understand the possibilities.

We haven't released our cloud platform yet, but if there are particular domains you are interested in, you can poke around http://reference.wolfram.com/language/ to see what we have there. The examples to do with machine learning seems to be popular on http://www.wolfram.com/language/gallery/.

Then teach me. I have no idea what to expect from Alpha, and I've so far only used it as a novelty item or advanced calculator. You can't expect me to 'naturally' speak a new language if you never reveal its vocabulary or grammar. And I can't learn the language intuitively as long as I can't hear it spoken anywhere either.

"apply color blindness to photo of ron howard" is also a valid query

So I tried "apply color blindness to photo of jonti roos" but got no useful results. However, "apply color blindness to http://assets.nydailynews.com/polopoly_fs/1.1717717.13945547... worked.

Original: http://assets.nydailynews.com/polopoly_fs/1.1717717.13945547...

I tried that, the the photos look identical. Of course I have red-green colorblindness.

do you know how to apply both color blindness and dog vision?

"Apply mantis shrimp vision" would be more impressive.

But this doesn't work ?

"apply cat vision to photo of obama"

Cat vision isn't implemented it seems.

But this doesn't work: "apply dog vision to a photo of some guy" http://www.wolframalpha.com/input/?i=apply+dog+vision+to+a+p...

http://www.wolframalpha.com/input/?i=apply+dog+vision+to+pho...

I like this one better:

http://www.wolframalpha.com/input/?i=apply+dog+vision+to+pho...

"Wolfram, could I see a hat wobble and a flarhgunnstow?"

"Yes Paul"

We have important work to do.

"apply dog vision to a photo of a Michael Jackson" really interesting! lol http://www.wolframalpha.com/input/?i=apply+dog+vision+to+a+p...

Woofram Alpha.

I highly doubt that 20% is uniform across all breeds.

That's the problem with studying animals, they can't defend themselves to correct our hilarious over-generalizations and misrepresentations of their species.

I did a little research on how they determine the visual acuity of dogs. Apparently it is done by training the dog to respond to lines on a background in order to receive a treat. Once the lines become so small that the dog no longer responds they have arrived at their answer.

What strikes me as odd is that they have taken a test designed for humans and modified it to be used for dogs. One issue with this is the fact that when a person takes a vision test they are trying very hard to discern the correct letter. Anybody who has gone thorough one of these tests will know what I am talking about. At some point you get to a letter size where you will spend considerable effort scrutinizing it. Is it a 'C' or a 'G'?

It seems to me that you should probably design a test for dogs that brings out a similar level of effort.

fixed it: http://www.wolframalpha.com/input/?i=apply+dog+vision+to+pho...

You shouldn't trust that interpretation by Wolfram Alpha.

1. Distance. One obvious discrepancy is that the closer you look at the photo the less detailed it'll look. That's the opposite of how actual "vision" works. Even in dogs. You need to know how far away you should look at the photo, at what DPI screen in order for it to be even remotely representative for what it "feels" like for a dog to see.

2. Original image dpi. The second flaw is that if that image was a 600 dpi photo McNamara processed through the same filter, it wouldn't have lost nearly as much information percentage-wise, as a small image would.

3. What is "human vision"? The original image doesn't represent "human vision". It's a photo captured by a camera, resampled to a specific (small) resolution. Wolfram Alpha can't know how typical human vision "dpi" applies to that particular photo, because it needs to know a lot of information about the camera lens and sensor, which it doesn't know. So it can't approximate the dog version out of it by using relative human-to-dog metrics.

I love Wolfram & all their products.

However, sorry, but that "dog vision" image is not scientific. It's just bullshit.

Harsh! The 2 color receptor part is accurate. And given that both photos are on your screen, and one is 20% less acute than the other, the approximation is fair.

Also, a lot of what we think of as vision happens in "software." A human looking at a human processes that information differently to a dog looking at a human.

Human's read faces, for example. To a human it seems that a lot more information is lost in the from the face part of the image than the hands area when you put it through that filter.

Excellent points, and why is the resulting picture larger than the original source image?

Sounds like more BS from this company (I don't have a ton of experience with their products but have followed their promotional materials quite frequently.)

Is there a political message here? How McNamara is viewed by a dog. Does the dog represent soldiers of Vietnam? ex. dog tag. If so, does the lack of clarity show the failing of McNamara in the Vietnam war?

Sometimes a cigar is just a cigar.

like the one that almost blew up Castro

Who would make a political statement in some obscure internet page that .01% of users use, about a 50 year old war?

Reply on news.ycombinator.com