Artificial Intelligence meets Bonsai

[jooni]

https://imgur.com/EQ3TywL

(This video was automatically generated by an AI from a single 2D image.)

Hi all, I wanted to share a small experiment we've done. We applied an AI algorithm to automatically render 2D Bonsai images as 3D videos with pre-set camera movements in order to give a better sense of a tree's volume.

What do you think? Does it help anyhow or is only misleading?

(credits of the original photo: grobonsai)

 

[leatherback]

I would recommend NOT sending people off to a vague external website with lots of parameters, but uploading the video to a recognized platform. Too much phishing going on.

 

[jooni]

I would recommend NOT sending people off to a vague external website with lots of parameters, but uploading the video to a recognized platform. Too much phishing going on.

Hi, not sure if I got it, do you mean the fact that I posted the video on imgur? It should appear embedded in the post, no?

 

[leatherback]

strange. When I posted there was no video and only links to jooni.app

 

[jooni]

strange. When I posted there was no video and only links to jooni.app

Ohh ok, but does it show up now?

 

[Bonsai Nut]

I would recommend NOT sending people off to a vague external website with lots of parameters, but uploading the video to a recognized platform. Too much phishing going on.

Imgur is a recognized and approved media platform. There is no problem with posting media from there.

@jooni I think it is a really interesting experiment. Right now I'm not sure it is that useful, but I could easily see a future where you take four 2D images from all four side of a tree, and AI would stitch the images together into a 3D rotational model.

 

[sorce]

Welcome to 01100010100!

Sorce

 

[RJG2]

Welcome to 01100010100!

Sorce

**
01000011 01110010 01100001 01111010 01111001
**

 

[jooni]

**
01000011 01110010 01100001 01111010 01111001
**

01101000 01100001 01101000 01100001

 

[TomB]

.

 

[Wires_Guy_wires]

My AI has been sipping lemon tea with a scoop of sweets from the blue honey jar.

 

[Grovic]

https://imgur.com/EQ3TywL

(This video was automatically generated by an AI from a single 2D image.)

Hi all, I wanted to share a small experiment we've done. We applied an AI algorithm to automatically render 2D Bonsai images as 3D videos with pre-set camera movements in order to give a better sense of a tree's volume.

What do you think? Does it help anyhow or is only misleading?

(credits of the original photo: grobonsai)

At first it would seem like I'm looking a bit to the sides of the tree, but in reality, you are not showing much more information than what is shown in the 2D image.
Now, I know nothing about marketing, will the animated picture sell more items? If it does, I'd say it's just psychology.
Two pictures (front view, and one of the sides) would tell me more about the volume, and I assume that would be less work for you too.
An even better option would be a 3d model as suggested by @Bonsai Nut .

 

[Bonsai Nut]

At first it would seem like I'm looking a bit to the sides of the tree, but in reality, you are not showing much more information than what is shown in the 2D image.
Now, I know nothing about marketing, will the animated picture sell more items? If it does, I'd say it's just psychology.
Two pictures (front view, and one of the sides) would tell me more about the volume, and I assume that would be less work for you too.
An even better option would be a 3d model as suggested by @Bonsai Nut .

The problem with this AI projection is that it is not really AI - in terms of the fact that there isn't really an intelligence "guessing" at the unseen content in order to fill in a 3D image. If you look at it closely, it is actually smearing the information that it knows, instead of creating new information based on an advanced algorithm. If you were to provide the 2D image to a human, and ask them to create a 3D model, they would use their intelligence to guess at what lies behind the tree, without knowing it. (Including guessing what the fence behind the tree looks like, where the background landscape trees go, etc).

So by providing several images from different angles, you're providing more information to what the AI "knows", and based on current models, reducing the blackhole of what the model doesn't know.

However... in the future... real AI will take that first 2D image, evaluate the species, the branch structure, the age, the pruning... and will calculate what the back of the tree MUST look like. And will model it appropriately. That is the difference.

 

[HoneyHornet]

I would like to see something that takes a picture of a tree and does its AI business to it to show it as bonsai..if somehow it knew what cuts would be optimum and rhen show how those changes would age

It would be totally approximate but kind of neat.. Are there not programs to make a oocture look 3d..I have a droid but i thought iphones take pics like that..i like the idea tho..sometimes a really well focused shot can grab that too

 

[jooni]

The problem with this AI projection is that it is not really AI - in terms of the fact that there isn't really an intelligence "guessing" at the unseen content in order to fill in a 3D image. If you look at it closely, it is actually smearing the information that it knows, instead of creating new information based on an advanced algorithm. If you were to provide the 2D image to a human, and ask them to create a 3D model, they would use their intelligence to guess at what lies behind the tree, without knowing it. (Including guessing what the fence behind the tree looks like, where the background landscape trees go, etc).

So by providing several images from different angles, you're providing more information to what the AI "knows", and based on current models, reducing the blackhole of what the model doesn't know.

However... in the future... real AI will take that first 2D image, evaluate the species, the branch structure, the age, the pruning... and will calculate what the back of the tree MUST look like. And will model it appropriately. That is the difference.

It may not be guessing much but would still be super useful I think, it could also help with choosing a front for the trees

 

[Harunobu]

I assume this is a deep learning technique. Whether that qualifies as 'AI' or not is a philosophical discussion. Yes, it is 'smearing' information from previously seen pictures unto your (bonsai) picture. And how well it does that depends on how well trained it is.

There are some nice coloured 60fps HD videos now of very old black&white footage. It is a bit tricky to interpret what you are seeing. Is it an accurate representation of what is actually there? Or just one possible option out of many? And what kind of artifacts, biases, and loss of detail is introduced that is not noticeable?
Some historians have gone on a bit of a rant against these unscaled videos. And these historians are kind of making silly and emotional arguments that seem to indicate they misunderstand the technique used or the reason people like to watch them. Things like saying "colourisation does not bring us closer to the past; it increases the gap between now and then. It does not enable immediacy; it creates difference."

It is like saying that the 3d representation of a bonsai removes us further from the actual bonsai. But the question is how accurate the 3d representation is. It is not real. But in what way? How much information do you need to know how a bonsai would look in 3d?

In the end any form of intelligence is 'just an algorithm', including our brains. So just saying that doesn't mean much to me.

 

[Potawatomi13]

Imgur is a recognized and approved media platform. There is no problem with posting media from there.

@jooni I think it is a really interesting experiment. Right now I'm not sure it is that useful, but I could easily see a future where you take four 2D images from all four side of a tree, and AI would stitch the images together into a 3D rotational model.

Would this be same thing as Alvaro uses showing his sale trees rotating?

 

[Potawatomi13]

I would like to see something that takes a picture of a tree and does its AI business to it to show it as bonsai..if somehow it knew what cuts would be optimum and rhen show how those changes would age

It would be totally approximate but kind of neat.. Are there not programs to make a oocture look 3d..I have a droid but i thought iphones take pics like that..i like the idea tho..sometimes a really well focused shot can grab that too

This completely removes instinctive gut feelings or visions used as individual designs/decisions made to develop a tree. Much like developing by committee design likely turns out bland and way too adherent to the "rules". Ultimately cookie cutter trees.

 

[HoneyHornet]

This completely removes instinctive gut feelings or visions used as individual designs/decisions made to develop a tree. Much like developing by committee design likely turns out bland and way too adherent to the "rules". Ultimately cookie cutter trees.

Yea i get that and agree, I was only trying to give life to an idea that i thought needed some oomph in the innovation dept