"Example-Based Sampling with Diffusion Models. (arXiv:2302.05116v1 [cs.GR])" — A generic way to produce 2-d point sets imitating existing samplers from observed point sets using a diffusion model which addresses the problem of convolutional layers by leveraging neighborhood information from an optimal transport matching to a uniform grid, that allows benefiting from fast convolutions on grids, and to support the example-based learning of non-uniform sampling patterns.

Paper: http://arxiv.org/abs/2302.05116

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Learning Rank-1 realizations us…

The images show two separate iterations of the same app — the first is a SwiftUI app for doing #StableDiffusion image generation using #CoreML. That took several weeks of work to create and it never really worked right — I could only do single selection of images, couldn’t drag and drop more than one image, and it took a fair amount of time to implement even trivial stuff.

The second is the same app implemented using AppKit. It took essentially one day (yesterday). Multiselection was built-in, drag and drop (for multiple items) was a couple of lines of code, and I implemented a bunch of other things I wanted to do fairly easily.

On top of that, the app feels faster, lighter, and more responsive than the SwiftUI version for the exact same task, using the exact same models.

SwiftUI has a long way to go still, if it will even ever get there …

#Coding #macOS #Swift #SwiftUIvsAppKit
Screenshot of an image generati…
Screenshot of an image generati…

Opera is going to use #ChatGPT to shorten articles — so that people have to read less and think even less —, Microsoft is building ChatGPT into their Office tools — again, less thinking —, we already have ChatGPT in search engines ..

So this is how the world ends — everybody growing stupider because we are too lazy to think for ourselves, and not with a bang?

I guess SkyNet did arrive but not in the way we thought … 😛

#TheAIRevolution #MisinformationEngine
Engadget article — Opera is add…

For those of us outside the Dole/Chiquita hegemony, bananas can have a variety of tastes — sweet, sour, bland and all the other stuff in between 😛 So when you say “water bugs taste like banana”, which banana are we talking about?

https://en.wikipedia.org/wiki/Banana
Chefs who cook with insects rep…

I really miss having @qikipedia posting facts over here. For a week or so, they did and then they’ve gone back to only posting on Twitter 😞

So I guess I’ll have to repeat their facts since I do enjoy them … but then there is no quoting posts and I have to resort to screenshots .. Ah, well 🙂
There is a termite city underne…

"UniPC: A Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models. (arXiv:2302.04867v1 [cs.LG])" — A unified corrector (UniC) that can be applied after any existing DPM sampler to increase the order of accuracy without extra model evaluations, and derive a unified predictor (UniP) that supports arbitrary order as a byproduct.

Paper: http://arxiv.org/abs/2302.04867
Code: https://github.com/wl-zhao/UniPC

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
The main idea of UniPC. We prov…

"Trading Information between Latents in Hierarchical Variational Autoencoders. (arXiv:2302.04855v1 [stat.ML])" — A generalization of VAEs to application domains beyond generative modeling (e.g., representation learning, clustering, or lossy data compression) by introducing an objective function that allows practitioners to trade off between the information content ("bit rate") of the latent representation and the distortion of reconstructed data.

Paper: http://arxiv.org/abs/2302.04855

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Left: trade-off between perform…

"Robot Synesthesia: A Sound and Emotion Guided AI Painter. (arXiv:2302.04850v1 [cs.CV])" — An approach for using sound and speech to guide a robotic painting process by encoding the simulated paintings and input sounds into the same latent space.

Paper: http://arxiv.org/abs/2302.04850
Code: https://github.com/pschaldenbrand/Frida

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Our robot painting sound guidan…

"Is This Loss Informative? Speeding Up Textual Inversion with Deterministic Objective Evaluation. (arXiv:2302.04841v1 [cs.CV])" — A study of the training dynamics of textual inversion, with the aim of speeding it up.

Paper: http://arxiv.org/abs/2302.04841

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
A summary of our key findings: …

"Better Diffusion Models Further Improve Adversarial Training. (arXiv:2302.04638v1 [cs.CV])" — Tries to answer the question "Can better diffusion models further improve adversarial training?" empirically.

Paper: http://arxiv.org/abs/2302.04638
Code: https://github.com/wzekai99/DM-Improves-AT

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
A brief summary comparison of t…

"Adversarial Example Does Good: Preventing Painting Imitation from Diffusion Models via Adversarial Examples. (arXiv:2302.04578v1 [cs.CV])" — Exploration into creating an image that is similar to another image for human vision, but unrecognizable for diffusion models to investigate the possibility of creating images that hinder diffusion models from extracting their features.

Paper: http://arxiv.org/abs/2302.04578

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Comparison of workflows for adv…

"Efficient Attention via Control Variates. (arXiv:2302.04542v1 [cs.LG])" — A look at control variates to show that Random-Feature-based Attention (RFA) can be decomposed into a sum of multiple control variate estimators for each element in the sequence.

Paper: http://arxiv.org/abs/2302.04542

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Left and middle: empirical memo…

"Feature Likelihood Score: Evaluating Generalization of Generative Models Using Samples. (arXiv:2302.04440v1 [cs.LG])" — A parametric sample-based score that uses density estimation to quantitatively measure the quality/diversity of generated samples of deep generative models while taking overfitting into account.

Paper: http://arxiv.org/abs/2302.04440
Code: https://github.com/marcojira/fls

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
FLS for a generative model that…

"Q-Diffusion: Quantizing Diffusion Models. (arXiv:2302.04304v1 [cs.CV])" — A solution to the slowness in diffusion models to generate images by compressing the noise estimation network to accelerate the generation process using post-training quantization (PTQ).

Paper: http://arxiv.org/abs/2302.04304

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Stable Diffusion 512 ×512 text-…

Thought I’d put the new #ChatGPT plugin through its paces (I’m not linking to it, given the results, I think I’d be doing everybody a disservice linking to it …) and installed it. The very first Google query I had was for something coding related that I was stuck on …

The ChatGPT answer? Looks convincing but total BS 😛

PSA: Kids, do not rely on ChatGPT and always verify the info you receive from strangers (and ChatGPT or its ilk).

#MisinformationEngine #ChatGPT #Search
If you're using a `LazyVGrid` i…

"What do we learn? Debunking the Myth of Unsupervised Outlier Detection. (arXiv:2206.03698v2 [cs.CV] UPDATED)" — An investigation into what Auto-Encoders (AE) actually learn when they are posed to solve two different tasks and challenging the assumption that AEs are likely to be even better at reconstructing some types of Out-of-Distribution (OoD) samples.

Paper: http://arxiv.org/abs/2206.03698

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Unsupervised outlier detection …

"Improved Vector Quantized Diffusion Models. (arXiv:2205.16007v2 [cs.CV] UPDATED)" — Two techniques to improve the sample quality of VQ-Diffusion. 1) Explore classifier-free guidance sampling for discrete denoising diffusion models to propose a more general and effective implementation of classifier-free guidance. 2) Use a high-quality inference strategy to alleviate the joint distribution issue in VQ-Diffusion.

Paper: http://arxiv.org/abs/2205.16007
Code: https://github.com/microsoft/vq-diffusion
Demos: https://huggingface.co/spaces/patrickvonplaten/vq-vs-stable-diffusion
https://huggingface.co/spaces/williamberman/vq-diffusion

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
In-the-wild text-to-image synth…

“Mining Effective Features Using Quantum Entropy for Humor Recognition” — A paper which looks at dividing a joke into two components — the setup and the punchline — and examining the possible semantics and relationships between these components.

Paper: https://arxiv.org/abs/2302.03716

#NewPaper #Language #Computation

<<Find this useful? Please boost so that others can benefit too 🙂>>
A humor and non-humor example c…

"Zero-shot Generation of Coherent Storybook from Plain Text Story using Diffusion Models. (arXiv:2302.03900v1 [cs.CV])" — A neural pipeline for generating a coherent storybook from the plain text of a story by leveraging a combination of a pre-trained Large Language Model and a text-guided Latent Diffusion Model to generate coherent images.

Paper: http://arxiv.org/abs/2302.03900

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Zero-shot generation example of…

"Neural Artistic Style Transfer with Conditional Adversaria. (arXiv:2302.03875v1 [cs.CV])" — Two methods that step toward a style image independent neural style transfer model which could generate semantically accurate generated images for any content, style image input pair.

Paper: http://arxiv.org/abs/2302.03875
Code: https://github.com/nipdep/STGAN

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
The generated image (c), which …