Uh-oh:
Spiraling down the #nixOS road.

After watching this video (www.youtube.com/watch?v=dsl_...), I thought I should also have dinner. Dinner with the penguins and the snowflakes.

Let's see how far I am getting with this. Not sure I am ready.

Honestly flabbergasted by all the amazing work presented at #BCon25. It’s nice to see what everyone makes out of Blender and how far contributions can go. Definitely one of the nicest communities to be around. Superb talks, nice location, great people. Can’t wait to see you all next year again 🧡🎉

First @blender.org Conference in person. Let’s gooooooo!

It's a wrap! Thanks for the ride @blender.org!
I had a great time and learned a lot of stuff. Thanks to Omar and Habib for the mentoring, and I will definitely continue to contribute to Blender. :)

I'll be around at BlenderCon this week and happy to chat :)

I was today years old to discover that C++ also has:

- && `a and b`
- || `a or b`
- !a `not a`

Meet Higgins!
Not my goofball but a friend’s favorite 😬

A minimal example of how to build your scene with MSD to render: 昇る太陽

Not sure how many will affect this, but I assume this will kill a lot of the rebuttals. #neurips

@wetafxofficial.bsky.social and James Cameron bring us the next sequel of #Avatar to the theatres coming this December 19th! Get ready to experience the Ash people.

To whom it may concern

First @blender.org contribution merged into main! 🥳
It's a minor one, but a start for many more to come!

Please go ahead and read the full reply.

I would suggest to not just quote a reply without setting any context. Thanks.

Further it is theft, because the training data has not been eradicated and the model has never been put down. Commercially use without consent of the artists.

Hao & Romero 2024 :: Meshtron
Looks pretty sick.

Paper: arxiv.org/abs/2412.09548

Check their project's website: research.nvidia.com/labs/dir/mes...

(Video belongs to the original authors and is shortened for upload reasons).

Images are encoded using DINO and decoded from images onto the triplane. Different viewports and respective features are fed through multiple MLPs (SDF, Deformation, and Weight will be propagated to FlexiCubes). With the Albedo MLP, the network predicts a mesh.

Ge & Lin 2024 :: Photometric Stereo Based Large Reconstruction Model

arxiv.org/pdf/2412.07371

PRM is a photometric stereo scene reconstruction model based on a two-stage optimization using InstantMesh. This optimization utilizes triplanes and volume rendering and is followed by FlexiCubes.

The authors assume images are puzzles made of many pieces. Reference and test images are embedded using a SqueezeNet. Next, compute the cosine similarity between layers 2-4 image patches and build a similarity map for each image. The similarity map is re-scaled to the image's original dimensions.

Another day, another paper.

Hermann et al. 2024 :: Puzzle Similarity: A Perceptually-guided No-Reference Metric for Artifact Detection in 3D Scene Reconstructions

arxiv.org/abs/2411.17489

The authors propose a new no-reference metric for 3D scene reconstruction artifacts & an annotated dataset.

Fresh outta Vision Science x Computer Graphics preprint press :: Perceptually Optimized Super Resolution (Karpenko et al. 2024)

arxiv.org/abs/2411.17513

Dynamically-guided SR to human-sensitive areas in an image leading to half the flops used for an indistinguishable result from prior methods.