Hacker News Reader

Comment by foobarbecue | Hacker News Reader

The idea that the image comes in "upside down" has always seemed incorrect to me. Isn't it more likely that this is essentially a configuration variable that is initialized undefined, and then set correctly based on comparing the input to other systems (inner ear etc.)? Is there any evidence that it is ever set "incorrectly"? i.e. a baby reaching up instead of down for something?

Edit: to clarify, I'm saying I don't see any reason to think anything in the vision system ever "expects" light from the world's up direction to come in at the top of the retina. Surely the developing brain doesn't know what orientation or flip to expect, and the whole system is constructed based on initial inputs. It was never upside-down.

View on HN

If you wear glasses that invert the image, your eyes will adapt to the image after a few days. There is some information on the Wikipedia page of the guy who discovered it (https://en.wikipedia.org/wiki/George_M._Stratton).

Yes, I know. That's why I called it a configuration variable.

I think it has more to do with the content itself, and the elasticity and adaptability of neural networks. There is no "up" or "down" with inputs. I learned this when working with a machine vision problem. A model will adapt to the orientation or format of a given input. The orientation itself is never a variable, the network simply doesn't care and adapts to it. If you train a model to recognize input in a given orientation, any other orientation will give a high error rate. If you train the model with _every_ orientation, it will be able to classify inputs that are upside down, sideways, mirrored, etc., all relative to the original orientation.

Exactly. There is no "upside-down" unless you train one way and then flip it. So I wish people would stop saying "babies see upside-down at first."

To your point ... FWIW, young kids in cultures where they were carried in all different orientations as infants allegedly would not turn over a picture handed to them "upside down" to look at it - whereas kids who were held more upright as infants would. Presumably the former could make sense of objects in many more orientations. (Somewhat sure this was in Ashely Montegue's treatise on Touch.)

As for 'errors', yes, they are typically signs of trauma or dysfunction along the neurons that go into V1. You can have people that have blindsight or that can't tell faces apart in the upper half of their vision. You can get some pretty tragic stuff.

You can also have glaucoma or other 'typical' diseases of the eye. Things like blindspots from welding or lasers.

The visual system in healthy people is already a bit odd. You have the foveal blindspot, the differing distribution of rods and cones, the unseeing neurons that do light sensation for melanin production, etc.

Really, the whole sensory chain is one giant bag of 'wait, it does that? huh?'

https://en.wikipedia.org/wiki/Riddoch_syndrome

Above comment misplaced?

I had an eye surgery for strabismus when I was 5 or 6. They tightened some of the eye muscles. I was seeing everything rotated initially, and over a few weeks the angle gradually went from about 30 degrees back to zero.

I guess it was because the muscle adjustments resulted in my eyeballs getting rotated.

Sometimes I wish I could just forget about all the bizarre things our bodies do.

Yeah, ditto for glasses that change/remove colors ... after I few days your perception adapts and you see colors normally again.

It all proves how we don't directly see what's coming thru our eyes - we see the mental model that our brain maintains about the outside word, built/updated via what's coming in thru our eyes.

Please hint me to a source on this. I wonder since some time what will happen when you use a VR headset with see-trough cameras and change the color mapping.

Try warm-tinted sunglasses for a few hours - world becomes blue for a minute or two when I take mine off. I thought everyone noticed this at some point. I would be surprised if you couldn't replicate it with a VR headset, although that doesn't sound very comfortable. You would probably also experience a fleeting feeling of hyperrealistic vision, given how low-fi the see-through camera would be compared to your eyes.

I read this a long time ago .. don't remember the specific source. If I google for it I'm finding "The Innsbruck Goggle Experiments", but not sure if this was the same study I've seen before.

https://pubmed.ncbi.nlm.nih.gov/28521154/

http://users.sussex.ac.uk/~ezequiel/AS/Kohler_1962.pdf

What I've always wondered about this is what the experience is like during the transitionary period. Surely it doesn't instantly flip from completely upside-down to completely upright.

Another question I had is could you learn to instantly switch between them? If you alternated between wearing the glasses and not every few days, would the adaptation period shorten each time? Would it eventually become instant? This is usually the case for motor skills e.g. controlling something with an inverted vs. non-inverted axis.

Yes. It was never upside-down.

The fact the cells are not aligned with our visual representation doesn't mean it's "upside-down". Like you said, it's undefined, and the brain figures out what these cells represent.

There's no flipping going on.

Rays of light travel in straight lines. Light enters your eye and hits the back of your retina inverted, not just upside down. Left is right. Up is down.

[1] https://www.schoolphysics.co.uk/age11-14/Light/text/Pinhole_...

Right, but how is this any different than a camera? Nobody goes around saying Nikon designed their cameras upside down and then had to hack around the mistake in software. It's just how lenses work.

> Right, but how is this any different than a camera?

The commenter above argues the image isn't upside down.

> Nobody goes around saying Nikon designed their cameras upside down and then had to hack around the mistake in software. It's just how lenses work.

That's an oddly specific statement. A camera inverts an image using any combination of lenses, mirrors, and software. And if you were building an eye, you would be meticulously poring over these implementation details and discussing all your options. Should you revert the image by adding another lens? Or, do an inverse affine transformation? Maybe you can cross the optic nerves so that the brain gets the correct image? Or program a reward function to individual neurons guided by the vestibular system... etc.

The screen you are looking at right now has gone through at least two transformations - due to the physical nature of light. The fact that you can look around and not realize all this is happening is pretty amazing. There are a few leaky abstractions as noted in the twitter thread.

No - if you were creating a lens or sensor you wouldn't even think about the orientation of the image relative to the real world. It's irrelevant - only that the pixels can be read off in an orderly manner. This is even true for analogue processes (but if the method of converting negatives to photos was different I can see it'd be a factor).

Maybe people in this thread are talking past each other but imo I agree with the commenter arguing that the image is ever "upside down", there is no image.

Because our brains form true stereo vision processing which requires two images that are combined into one.

Digital cameras do not do that, it's ho we get our true depth perception among other features.

As bit of trivia I’ve had procedure of laser ablation performed on one of my eyes upper edge of retina that detached. I’ve pink cloud from bleeding obfuse botton half of image in that eye for short time.

It is upside down in the sense that if you built a lens system like the eye, the image produced would be upside down.

The system evolved that way so it is never set "incorrectly" or "expected" to be the other way.

But that's kind of the point -- the orientation of the image on the retina is irrelevant, unless it's being looked at by a little man sitting right-side-up inside your head.

Fortunately, his eyes have their own set of lenses that flip the image the right way up again.

Reply on news.ycombinator.com