Chris Offner

CompSci MSc student at ETH Zurich, recovering filmmaker and 3D/VFX artist.
Visual and Interactive Computing & Machine Learning

2024-06-17

Would an Apple Vision Pro bought in Germany be usable without restrictions in Switzerland, with a Swiss Apple ID? Unfortunately 🇨🇭 is not among the 28 countries where the Vision Pro launches in July. Would Vision Pro apps simply be unavailable in the Swiss App Store?

#AppleVisionPro #Vision_Pro #AVP #VisionPro #AppleVision

2023-12-31

@j Yeah, that's the only reason I still use it. I hope it'll change but if the ML crowd hasn't made the switch by now, I'm not sure what it'll take to convince them. 😕

Chris Offner boosted:
Robert McNeesmcnees
2023-12-30

"To Read”

A screenshot of a desktop folder labeled “To Read - Recent Papers.” The folder is orange and has mathematical symbols on it. The background is a starry region of a Hubble photo.The warehouse from the end of “Raiders of the Lost Ark,” where the government loses objects it doesn’t want anyone to find.
2023-12-30

Microsoft’s Copilot app is now on iOS and it has all the polish and magic you expect from a Microsoft app.

This was the first prompt I tried, to see whether DALL-E 3 works without paying here, unlike in chatGPT where it’s a Pro feature.

#MicrosoftCopilot #LLM #AI #bingchat

2023-12-30

Hang this quote on every CV researcher’s wall.

#ComputerVision #ComputerGraphics #MachineLearning #ML #NeRF #GaussianSplatting

Chris Offner boosted:
Gokhan Avkarogullarigavkar@mastodon.gamedev.place
2023-12-30

@Blender blender continues to be a well run and impactful project. It is great to see @aras recognized. He is a terrific person and a great engineer that I look up to. Also good to see Jason Fielder and his team at Apple making a difference for the blender users.

Chris Offner boosted:
2023-12-29

The funny thing about LLMs is they’re not good for knowledge work because they sometimes make up stuff that doesn’t exist and they’re not good for creative work because they sometimes make up stuff that does exist.

2023-12-29

@stewartlynch8 They're still the result of an algorithm. The difference is just that the parameters used in that algorithm are learned from data rather than hand-coded.

Chris Offner boosted:
2023-12-29

Pouring one out for all my poor friends who have birthdays in the Dec 20 - Jan 4 dead zone when everybody's all partied out.

It's frankly just inconsiderate planning by parents. You've heard of no-nut-November, well how about we start No-Mother-March.

2023-12-29

@TomF My birthday is on December 24 and it's not the ideal date if, like me, you live in a country where Christmas Eve is a way bigger deal than Christmas Day.

2023-12-29

Here's a more clearly visible demonstration of the problem I described previously: sigmoid.social/@chrisoffner3d/

On the left we see the progression of cross-attention maps extracted via the CPU, on the right we see the same cross-attention maps extracted via the GPU.

This is using the #Keras implementation of #StableDiffusion on an M3 Max.

#TensorFlow #StableDiffusion #Diffusion #Python #MLEngineering #MachineLearning #DeepLearning #GPU #M3Max

2023-12-29

Cross-attention maps of resolution 64x64 in Stable Diffusion's U-Net for time steps t = 1, ..., 1000.

Chris Offner boosted:
2023-12-29

Something I am noticing a lot is how AI image generators have been finetuned to center subjects in the frame. But if you look at art you'll realize that there are so many ways to compose images that make them feel more dynamic. Center-weighted compositions are typically what beginners rely on. So I worry about the loss of variety when datasets are curated by novice artists or people without an appreciation of art and how this likely restricts a model's ability to create dynamic compositions.

2023-12-28

@at Haha, not at all – I'm actually very interested to learn more about it once this godforsaken semester is finally behind me. :)

Thank you! 😄

2023-12-28

@orsinium Thank you for the hint. I tried autopep8 but I vastly prefer black's style.

Chris Offner boosted:
2023-12-28

This bit is quite shocking
"Despite the flippant observation – often made by European officials – that Russia’s economy is the same size as that of Italy, the Kremlin is producing more shells than all of Nato. "

@guardian

Chris Offner boosted:
2023-12-28

Conda is moving our social media presence from Twitter/X to Mastodon and LinkedIn at the start of 2024. It's past time to move into spaces that are welcoming and more in line with our community values. Going forward, you can find us at
🐘 @conda (fosstodon.org/@conda) on Mastodon
🔗 Conda Community (linkedin.com/company/condacomm) on LinkedIn

Announcement: conda.org/blog/2023-12-27-soci
We hope to see you on Mastodon and LinkedIn in 2024!

Conda is moving to Mastodon and LinkedIn
Chris Offner boosted:
2023-12-28

As part of rewriting my hobby renderer to use Metal, I’ve now written a lot of supporting framework code to allow for easily cross-compiling complex code to both Metal and Cuda. Would this along with some sort of “Cuda programmer’s intro to Metal” be a useful post to anyone?

2023-12-27

@BartWronski Agreed, I think the highly idiomatic nature of the Chinese language is in tension with western academic style.

Client Info

Server: https://mastodon.social
Version: 2025.04
Repository: https://github.com/cyevgeniy/lmst