We address the long-standing problem of how to learn effective pixel-based image diffusion models at scale, introducing a remarkably simple greedy growing method for stable training of large-scale, high-resolution models. without the needs for cascaded super-resolution components. The key insight stems from careful pre-training of core components, namely, those responsible for text-to-image alignment {\it vs.} high-resolution rendering. We first demonstrate the benefits of scaling a {\it Shallow UNet}, with no down(up)-sampling enc(dec)oder. Scaling its deep core layers is shown to improve alignment, object structure, and composition. Building on this core model, we propose a greedy algorithm that grows the architecture into high-resolution end-to-end models, while preserving the integrity of the pre-trained representation, stabilizing training, and reducing the need for large high-resolution datasets. This enables a single stage model capable of generating high-resolution images without the need of a super-resolution cascade. Our key results rely on public datasets and show that we are able to train non-cascaded models up to 8B parameters with no further regularization schemes. Vermeer, our full pipeline model trained with internal datasets to produce 1024x1024 images, without cascades, is preferred by 44.0% vs. 21.4% human evaluators over SDXL.

OpenAI has added a new "outpainting" function to its text-to-image AI model DALL-E that lets the system generate new visuals that expand the borders of any given picture. In the example above, you can see how DALL-E, with the help of human prompting, "imagines" what's outside the frame of Johannes Vermeer's portrait "Girl with a Pearl Earring." Note, how, even from the limited information provided by the portrait, the system is able to match Vermeer's style, mimicking the shadows and highlights of the original. In the timelapse below, you can also see how the artist responsible, August Kamp, had to expand the image in small sections at a time, often redoing DALL-E's generations in order to get the outcome she wanted. Not seen in this video but definitely worth highlighting, is the fact that the system is not generating these extensions just by itself.

OpenAI has added a new "outpointing" function to its text-to-image AI model DALL-E that lets the system generate new visuals that expand the borders of any given picture. In the example above, you can see how DALL-E, with the help of human prompting, "imagines" what's outside the frame of Johannes Vermeer's portrait "Girl with a Pearl Earring." Note, how, even from the limited information provided by the portrait, the system is able to match Vermeer's style, mimicking the shadows and highlights of the original. In the timelapse below, you can also see how the artist responsible, August Kamp, had to expand the image in small sections at a time, often redoing DALL-E's generations in order to get the outcome she wanted. Not seen in this video but definitely worth highlighting, is the fact that the system is not generating these extensions just by itself.

Get access to everything we publish when you join VeloNews or Outside . Team Jumbo-Visma and its race clothing supplier AGU have created a limited-edition kit that will be worn during both the Tour de France and Tour de France Femmes. The new design was required because the team's typical yellow and black design is considered too similar to the maillot jaune worn by the race leader of the Tour de France. Last year the team created a bespoke kit for the Tour de France and the squad has followed suit for 2022. The Dutch team collaborated with AGU with the squad stating that its inspiration stemmed from Dutch artists Rembrandt Harmenszoon van Rijn, Johannes Vermeer, and Vincent van Gogh.

It seems there are few jobs robots can't do these days, even the most delicate jobs, like picking asparagus or potting plant seedlings. But they're only needed because humans can't - or won't - do the work, farmers say. Marc Vermeer had a problem. He was struggling to attract workers to pick his white asparagus crop in the Netherlands. The workers he did hire moved on quickly, so he was always training new people.

This repository contains a 1st place solution for the Painter by Numbers competition on Kaggle. Below is a brief description of the dataset and approaches I've used to build and validate a predictive model. The challenge of the competition was to examine pairs of paintings and determine whether they were painted by the same artist. The training set consists of artwork images and their corresponding class labels (painters). Examples in the test set were split into 13 groups and all possible pairs within each group needed to be examined for the submission.

Does every painter leave a fingerprint? Accurately distinguishing the artwork of a master from a forgery can mean a difference in millions of dollars. In the Painter by Numbers playground competition hosted by Kiri Nichol (AKA small yellow duck), Kagglers were challenged to identify whether pairs of paintings were created by the same artist. In this winner's interview, Nejc Ilenič takes us through his first place solution to this painter recognition challenge. His combination of unsupervised and supervised learning methods helped him achieve a final AUC of 0.9289.

Kathryn Hume bio, Director of Sales and Marketing, and Nick Vermeer (bio), Research Engineer, at Fast Forward Labs (helping organizations accelerate their data science and machine intelligence capabilities), join hosts Pavan Bahl, Rob Sanchez and Marc Raco. Hume and Vermeer discuss the genesis of Fast Forward Labs, how innovation changed what the Labs mean from then to now, natural language generation, learning how much distance there is to go with Artificial Intelligence and language, machine learning vs. AI, difficulties with Amy on scheduling and other platforms, a bot to mimic Alexis Ohanian, how much sample size matters, and image recognition. A conversation on use cases in fashion businesses, CRMs, collaborative filtering, a focus on brick and mortar, using satellite data for cars in parking lots and credit card activity, RFID and privacy laws, and the surprising best way to block an RFID signal. Attention directing research, the next frontier vs. solving today's problems, how Fast Forward labs is like an outsourced R & D department and mentor to grow internally, purple unicorns, being "nerd best friends", and a mechanism for applied research. Off the Grid Questions cover a death-inspired poem, jeweler screwdrivers vs. a computer, Long Island corn, crop rotation bar talk, reactive illuminated dance costuming, thermochromic paints, and fast jet skiing.