Alt text: In the 60s, Marvin Minsky assigned a couple of undergrads to spend the summer programming a computer to use a camera to identify objects in a scene. He figured they’d have the problem solved by the end of the summer. Half a century later, we’re still working on it.
Edit: seems I’m the third person to comment this! :')
I love how this is actually an example of progress. These days, ML can be used for this kinda thing and it’s not too bad at it even.
https://code.flickr.net/2014/10/20/introducing-flickr-park-or-bird/
This page about it still exists, but I guess the identification site died with Flickr.
“Cat? Dog? Pig? LOAF OF BREAD. SYSTEM ERROR!”
deleted by creator
Way back in the 90s I did a contract job at MS Research on a project called “V-Worlds” - a world simulator similar to the Doom or Quake engine, but it was browser-based and everything was a script, so changing how the world worked didn’t mean you had to restart a server, just change the scripts and new stuff would appear right in front of you.
Anyway the concept of adding accessories to the player’s avatar and even having a pet follow you around came up, and I remember there was an involved discussion of how difficult/impossible that would be. The player’s avatar was a special object class that represented a user, and didn’t have the same capabilities as ordinary objects in the world. I remember asking, “Why isn’t the avatar just a world object the player happens to control? Then you could do all kinds of cool stuff like let the player transform into something else just by switching objects, or let another player run your character.” Dead silence. I was just a contractor, what did I know?
This feels like the kinda project that should have a 1hr YouTube indie doc about it
I wouldn’t mind seeing that! After V-Worlds was declared “completed” MSR tried to find a product group to fold it into, but nobody wanted to own it. I don’t remember if XBox existed then, but the code just sat there for a few years, then I heard they opensourced it. When my kids were playing ToonTown I found a bug that let you slide behind the background and move around, like you could see that a clerk behind a counter was just a legless floating torso. The method of getting there seemed to be exactly like a V-Worlds bug, so I wondered if Disney might have been using the code. But it could have just been a common graphics bug, I dunno.
I remember finding another bug while creating a demo with a snaky sea creature swimming around. To animate a multi-segmented object you had to animate each segment separately. After the animation ran for a minute or two, enough unrelated interrupts would happen in the computer that would throw the body parts out of sync, making body parts either merge into each other or move apart, and the whole thing would look like crap. Same thing if you had somebody ride in a car or on a train - the car and character were animated separately and you’d end up with the character floating along behind the car.
I asked the dev about making the animation itself an abstract object whose position would be moved around, and attaching in-world objects to it, with position offsets. Each animation step would be computed just once instead of for each body part (or for the person and the car), and all the parts would be rendered with offsets from that one position, guaranteeing them to stay in sync visually. He said yeah that’s a good idea, but we’re not working on that code anymore. Oh well.
Another bug involved moving from room to room. The engine only loaded graphics for the current room, so when you went through a doorway it would load the new room and dump the previous one, causing a very unnatural visual delay that looked like a glitch in the matrix. The way we coped with this was by putting an entire world in a single room, so all the world’s graphics were loaded all at once. But this not only limited the world size, it meant we had to create our own version of the room system in script. To keep track of where players and objects were, we put invisible barriers in doorways and used event handlers when things passed through them. Then we used this to enforce which players could talk to each other or hear sounds made in a given “room”.
I suggested loading a cluster of rooms at once - the current one and those that were one connection away. Then when an avatar passed into a doorway the new room’s graphics would already be there, no glitch, and the graphics for nearby rooms could be loaded and unloaded in the background. Again, nice idea but we’re done working on that code. Sigh. I really wish I had joined that project about 6 months sooner. Not like I’m a genius or anything but these seemed like really fundamental things that should have been addressed up front.
Okay, rant over. I haven’t thought about this stuff in quite a while - I’m kind of amazed so many details are still in my head. I must have agonized over it a lot at the time lol.
Hey if it’s still out there as open source could finally fix those bugs lol!
But yeah seems interesting especially if it had a second life once it was opened sourced. It kinda boggles my mind how much companies are willing to scrap things after putting so much work into it, but I guess that’s the whole sunk cost thing but still.
Plenty of nostalgia for toon town so if it was used could broaden audience of video def.
Thanks for your ‘rant’ was interesting seeing some of the same problems that pop up for me in my current game dev (how to handle when to load certain things).
A browser based Doom or Quake engine world sim to run around playing with others sounds like such an awesome concept. I’d love that!! And in the 90"s no less. That would’ve been crazy impressive.
Microsoft and MMOs, man. I remember they were gonna make a really neat online fantasy one for the Xbox and canned it, too.
That’s such a wild story. Thanks for sharing that with us! I wish they wouldn’t have cold shouldered you like that…
Here’s how I was imagining that went down the whole time I was reading it lmao. Just for you.
Wow, nobody’s ever put me in a cartoon before. I’m honored!
It does kind of fit the situation lol.
There’s already a codebase for bursting from the ground in an explosion of lava. Everyone wants that.
You’re the first person asking for a scarf, and our system doesn’t even know what a neck is.
Time for the old NPC-with-a-train-for-a-hat trick.
Only in 3D. In 2D, you slap some pixels on top and there’s your scarf:
I tend to find it’s the other way around. Once you’ve got a scarf modelled and rigged, it’ll work* for all animations, but for animated 2D sprites you have a lot more things to do.
* May have visual artifacts like clipping
Do you have the Orb yet?
I actually do, yes. Hundreds, if not thousands, of hours played and I made it down there and back out exactly once. 🙃
Lol, I still died half way up my one orb run, so you’re further than me. I save scummed right before Zot:5 and did get it on my second try, but that doesn’t count.
It was a gnoll abyssal knight IIRC. What about you?
According to the screenshot I took, it was a gargoyle berserker with an axe. I had some ridiculous luck with armor drops, so basically every resistance was either maxxed or close to it. I only really got into trouble down in Zot:5…
And hmm, I should do more with Lugonu. I never really have a reason to pick him, but that means I also don’t experiment with him, so I won’t really learn what reasons there are to pick the guy…
Lugonu actually started as a goddess, before being turned into a genderless abomination. I think I’ve seen the devs use it/it’s pronouns.
Yup. The panic buttons are nice, the easier rune is nice, and back when there was Ak there was a cheap blink+blink other ability. That character used a distortion-branded demon whip to great effect as well, IIRC. Happy playing!
and add a couple of frames to the sprite sheet in order to animate the scarf if that’s required.
The giant is easy. The ground is easy. The lava though… Do you want the particles to stick together? To visually connect? To collide with each other? To interact with dynamic objects?
Design lead wants parting earth and flowing lava. Budget dictates static assets and baked in animations.
The player will jump on it, what happens then?
No problem, just mention it in the requirements - early on. Not when everything is built to work this one specific way.
Player? Easy. Scarf? Easy. Wearing a scarf? That depends on a lot of factors such as which part of the body, how the models were made and rigged, etc.
And if it like blows in the wind that’s a whole jigglebone system and wind simulation that’s a lot of stuff going on
I want dresses, and I don’t care if they clip through literally everything!
My bg3 character is female. She was in slacks until act 3 where she could finally have a dress
We looted everything. I feel like there are two dresses in the game: the robe Gale wears and a white dress you find in a Balders Gate house near the end of the game
mf said choas
Shadows in the real world a lack of energy Shadows in games imma need it all boss
“I noticed the elves in level 3 look too similar to the dwarves in level 5.”
“It’s too late to change it now!”
Sure. Player character? No.
That’s cute 😅
Tbf, you can make the characters wear anything, but it won’t look good. Lol
Welcome to second life
There’s an XKCD comic for that.
This comic is so old, that both should be rather easy now
It took almost exactly 5 years from publication for that to be commonplace.
TBF it had been a long standing problem for roughly a half century before this. Specifically birds were the thing researchers tried to identify first, which is probably the reference here.
She did get her research team after all :)
Now try to identify if it’s a fish
Oof
Fish? You mean a water bird?
🐬
There’s no such thing called fish.
- Stephen Jay Gould (Biologist)
Or there’s fish and we are one, or there’s fish but hagfish, dogfish and lungfish are something else.
I guess we could return to medieval and say it’s based on shape not taxonomy, too, so whales would be fish.
It’s why taxonomy uses latin for this… the definition of english words are based on common usage which isn’t going to line up to any kind of scientific categorization. English is always changing and scientific categorization is also always changing when there’s more empirical data. These changes are independent of each other so it was wisely decided long ago to not even try to make english words consistent with scientific taxonomy.
So in common usage, yeah it’s based around the general shape but it isn’t a whale (big mammal) a dolphin (a relatively smaller mammal). A shark might be called a fish but more likely someone will just call it a shark instead of just using just “fish”. This is fine for communication among laypeople, if marine biologists are having a conversation about those same animals, they break out the latin and there’s no confusion.
Also my understanding is that in medieval times, the word whale actually refereed to a specific species of whale… what we know call the Right Whale, which is nearly extinct. So a word for a species became a word for a group of species and then it was awkward how to refer to that original species. What kind of whale is that? “It’s a whale whale… you know the original whale… the proper whale… the right whale.” There’s actually a paragraph in Moby Dick about this.
English is weird and changes in weird ways. Just use latin if you want to be scientifically precise.
I mean, some of the taxanomic divisions do have common names as well - jawed fish and ray-finned fish might come up in that conversation. And don’t forget some of the formal names and roots are Greek as well.
What kind of whale is that? “It’s a whale whale… you know the original whale… the proper whale… the right whale.” There’s actually a paragraph in Moby Dick about this.
Was that the weird chapter that was just a biology lesson, but was also completely wrong?
Oh, yeah, the specific example listed was solved within roughly a month of the comic being posted. But the idea still applies, as seen with the twitter post above.
Well, sure, with an image classifier, the bird identification is doable. I’m sure I could implement that if I went looking for some open source thingamabob that does that. But it’s still not something I could actually understand. That part definitely hasn’t changed over the years.
Having taken an ML class, with some of my college notes I could do this and “understand” it… but the weights would still be a black box. AI training is black (box) magic.
Ebird/ Merlin bird id does this wonderfully.
only because people never stopped asking it to be able to id birds.
Can’t wait for the Disney live action movie remake of XKDC comics!
Plot twist: it will be written and created by AI:-(.
It will feature all sorts of jokes about using qwerty and windows
And Linux, if the Fediverse had any say about the matter! (Or xkcd either:-D)