#vectorised

[bruneburg]

Finally vectorised that fugitive carrot I doodled during a livestream last year.

[squirrellypoo]

It’s not 100% perfect but I wasn’t about to hand trace all those music notes (and after I spent ages tracing all the lettering I found good mimic fonts anyway 🙃).

But I’ve now got the S3 production art in vector format should I find a use case for it…

I don’t think I’ll cut this on heat transfer vinyl just yet as I still need to sew a shirt for the Paris Sucks motif (from that S3 press pack).

If anyone wants the svg vector file let me know!

[vreemd]

okay I'm high now and I'm having a drink I'm fiiiine. I'm so zen

[sapphicsundial]

Potential unpopular opinion but I hate the vectorised sapphic flag, the original with the illustrated violets are so much prettier. Violet petals aren’t even round like that.

Kind of reminds me of the two Venice flags. I like both of them but the one with the paintings is so much more opulent.

[aviyx]

im slowly but surely practicing being able to do less shakey lines on this tablet and its going well :)

currently working on those stickers i said i was doing, and lets just say i adore these (even though the lineart on the 1st one look me literally 5 hours because draw undo draw undo etc etc)

[letiziacasanova-tshirts]

Comment créer ses graphiques – Létizia Casanova (leticasanova.fr)

[neuromantic1]

Fine-tuning me-bot cost £11,550

Mad story I found on a Google Developer forum. Student tries to train AI-bot of self, for partner, and gets landed with £10k+ bill. Overnight.

It's true, Google Cloud Platform (databases, serverless hosting etc) is often derided for it's overly complex pay-as-you-go pricing. A lesson for us all.

An organisation or individual with clout could investigate this story. Get this student a break. Their (Chinese TikTok) Red Note story had a million views.

Read full story

Last week I tried something fun and romantic: I wanted to give my boyfriend a special birthday surprise, so I fine-tuned a large-language model with Google Vertex AI using our past chat history. The goal was for the model to know my memories and being able to reply exactly like a digital version of me.

I ran the job for about four hours at night and was confident it’d stay under the free USD $300 credits. I even set a £5 budget alert. The next morning, a few hours after waking up, I opened an email titled “Issue with your Cloud Billing” → £11,550 (€13,600 / $15,500) due.

As a full-time student living on a €425 monthly stipend (after rent I have €212 for everything), this was devastating. I thought the training would cost maybe $150-200. Instead, one short run got me a charge worth years of living expenses.

[gramatter]

Costs £1 per image vectorised

[valheyrie-v404]

Démarrez votre projet avec une image parfaitement nette et adaptable, idéale pour le web ou l'impression. En tant que créative spécialisée en création d'images par IA et débutante en vectorisation, je vous offre un service unique qui inclut gratuitement des variations générées par IA de votre image originale. Ces aperçus vous permettront de voir différentes interprétations créatives avant de finaliser votre choix pour la vectorisation.

Comment ça marche ?

Envoyez votre image : Partagez avec moi l'image que vous souhaitez transformer en vectoriel. Versions préliminaires par IA : Recevez gratuitement jusqu'à trois variations générées par IA de votre image originale pour explorer différentes options stylistiques. Choix de la version à vectoriser : Sélectionnez la variation que vous préférez pour la conversion en vectoriel. Création et Révisions : Comme je perfectionne mes compétences en vectorisation, je réalise la vectorisation de votre choix et offre jusqu'à deux révisions gratuites pour peaufiner le résultat. Livraison finale : Recevez votre image vectorielle en formats AI, EPS, JPEG, et PNG, prête à être utilisée dans tous vos projets. Pourquoi choisir ce service ?

Flexibilité et Personnalisation : Profitez des variations créatives autour de votre image initiale pour assurer que le produit final correspond exactement à vos attentes. Qualité et Adaptabilité : Les graphiques vectoriels sont ajustables à n'importe quelle taille sans perte de qualité, idéaux pour une utilisation professionnelle. Soutien à un artiste émergent : En choisissant mon service, vous soutenez mon développement professionnel dans le domaine de la vectorisation. Ce que vous recevez :

Fichier vectoriel original en AI. Fichier EPS pour une compatibilité maximale. Fichiers JPEG et PNG pour une visualisation facile. Droits d'utilisation complets pour vos projets personnels et commerciaux. Délai de réalisation :

Les commandes sont généralement complétées sous 7 à 10 jours ouvrables, ce qui permet de garantir la qualité et d'apporter les ajustements nécessaires.

Tarifs :

Vectorisation avec options IA : 25€ (inclut trois variations préliminaires générées par IA et deux révisions gratuites). Commandez maintenant pour transformer et personnaliser vos images de manière créative et précise ! 🌟 https://valheyrie404.etsy.com/fr/listing/1722212511/service-de-vectorisation-dimage-avec

[clementtisne]

Simplification d'image de film à la façons de Tom Purvis.

Vectorisation d'un passage du film Passengers.

(Illustrator)

[helyannis]

before this post by @eilinelsghost I somehow never realised that I could use MY ACTUAL REAL-LIFE WORK for something more fun??

so I threw the official maps into a geographic information system and aligned them as best I could (grid in km)

And I very roughly vectorised some of the woods and rivers of Beleriand so I could measure them more easily:

Looks like I managed to get the size reasonably accurate. We're told in the Quenta that the length of Sirion from the Pass to the Mouths is 121 leagues (= 584.2 km) and that it's 40 leagues (= 193.1 km) from the confluence of Little & Greater Gelion to the confluence of Ascar and Gelion.

Only Narog should be 80 leagues (= 386.2 km) long but is >50 km longer than that for some reason …

(ignore the ellipsoidal lengths. we cannot use a map projection of course)

SO NOW I can tell you fun things like that the Isle of Balar is almost as large as Mallorca (3109 km²) and the Forest of Region is about the size (and shape) of South East England (19 144 km²)!

if anyone needs measurements or wants the geopackage, feel free to ask!

[bruneburg]

A little beakhead wizard, full of ideas and potential.

[squirrellypoo]

I made an Interview with the Vampire font

As testement to my mental illness and 30+ year obsession with these vampires, I went and made a font from the lettering used in the tv show.

I've never made a font before, but I'm proficient in Adobe Illustrator, so after hunting down all the letters in the credits (in both weights!), I hand traced all the glyphs, and aligned and trued the various points. That actually wasn't too bad - the more time consuming part was creating from scratch all the symbols and numbers that weren't included in the credits at all (and I'm pretty happy with everything but the @ symbol tbh!)

It's a captal-only font, with the default being the thick version, and if you install the additional RueRoyale-Thin.otf, you'll get the thinner version in all caps, too.

Because I had a bunch of the ornate lettering and the little Immortal Universe logo already vectorised for vinyl cutting, I added those in as easter eggs, too. I've tried my best to included all the accented upper case characters for the foreign fans, but I've probably missed something, and I had to stop somewhere... If you see something horribly wrong though, please tell me in the comments.

You can download the font (2x .otf files, at under 1MB total) here.

This is a fan-created font, for fans to use FREE OF CHARGE, so I'm hoping that keeps AMC's lawyers off my tail. If you use it and want to credit it, please link back to the original tumblr post here in case I need to make any updates on it. Enjoy! ❤️

[bayesic-bitch]

Old-school planning vs new-school learning is a false dichotomy

I wanted to follow up on this discussion I was having with @metamatar, because this was getting from the original point and justified its own thread. In particular, I want to dig into this point

rule based planners, old school search and control still outperform learning in many domains with guarantees because end to end learning is fragile and dependent on training distribution. Lydia Kavraki's lab recently did SIMD vectorisation to RRT based search and saw like a several hundred times magnitude jump for performance on robot arms – suddenly severely hurting the case for doing end to end learning if you can do requerying in ms. It needs no signal except robot start, goal configuration and collisions. Meanwhile RL in my lab needs retraining and swings wildly in performance when using a slightly different end effector.

In general, the more I learn about machine learning and robotics, the less I believe that the dichotomies we learn early on actually hold up to close scrutiny. Early on we learn about how support vector machines are non-parametric kernel methods, while neural nets are parametric methods that update their parameters by gradient descent. And this is true, until you realize that kernel methods can be made more efficient by making them parametric, and large neural networks generalize because they approximate non-parametric kernel methods with stationary parameters. Early on we learn that model-based RL learns a model that it uses for planning, while model free methods just learn the policy. Except that it's possible to learn what future states a policy will visit and use this to plan without learning an explicit transition function, using the TD learning update normally used in model-free RL. And similar ideas by the same authors are the current state-of-the-art in offline RL and imitation learning for manipulation Is this model-free? model-based? Both? Neither? does it matter?

In my physics education, one thing that came up a lot is duality, the idea that there are typically two or more equivalent representations of a problem. One based on forces, newtonian dynamics, etc, and one as a minimization* problem. You can find the path that light will take by knowing that the incoming angle is always the same as the outgoing angle, or you can use the fact that light always follows the fastest* path between two points.

I'd like to argue that there's a similar but underappreciated analog in AI research. Almost all problems come down to optimization. And in this regard, there are two things that matter -- what you're trying to optimize, and how you're trying to optimize it. And different methods that optimize approximately the same objective see approximately similar performance, unless one is much better than the other at doing that optimization. A lot of classical planners can be seen as approximately performing optimization on a specific objective.

Let me take a specific example: MCTS and policy optimization. You can show that the Upper Confidence Bound algorithm used by MCTS is approximately equal to regularized policy optimization. You can choose to guide the tree search with UCB (a classical bandit algorithm) or policy optimization (a reinforcement learning algorithm), but the choice doesn't matter much because they're optimizing basically the same thing. Similarly, you can add a state occupancy measure regularization to MCTS. If you do, MCTS reduces to RRT in the case with no rewards. And if you do this, then the state-regularized MCTS searches much more like a sampling-based motion planner instead of like the traditional UCB-based MCTS planner. What matters is really the objective that the planner was trying to optimize, not the specific way it was trying to optimize it.

For robotics, the punchline is that I don't think it's really the distinction of new RL method vs old planner that matters. RL methods that attempt to optimize the same objective as the planner will perform similarly to the planner. RL methods that attempt to optimize different objectives will perform differently from each other, and planners that attempt to optimize different objectives will perform differently from each other. So I'd argue that the brittleness and unpredictability of RL in your lab isn't because it's RL persay, but because standard RL algorithms don't have long-horizon exploration term in their loss functions that would make them behave similarly to RRT. If we find a way to minimize the state occupancy measure loss described in the above paper other theory papers, I think we'll see the same performance and stability as RRT, but for a much more general set of problems. This is one of the big breakthroughs I'm expecting to see in the next 10 years in RL.

*okay yes technically not always minimization, the physical path can can also be an inflection point or local maxima, but cmon, we still call it the Principle of Least Action.

[insaturnity]

sanders sides MORE pins AND BRACELETS!!

patton stuff is fot @apairofmismatchingsocks , and the roman / remus stuff is for @jupiterrr33 !! anything virgil or janus related is for me hehee

also i did want to pt monster tabs on the bracelets but i dont have any blue ones, as we used them for our rainbow monster tab ..thing..?? on our backpack xD

and dude. i traced over a BLURRY ASF image of remus' patch and i looked him up today and found a VECTORISED ONE. .. AUHHHHGGHHAGAAAA

[vampyre-bytess]

So I’m going to be heatpressing these onto my little doll clothes for Cas so I figured I’d share them!

I vectorised these guys so they are clear (even though they will be tiny LMAO)