Pixel-Space 👏

Preparing a major PR to improve @ThreeJS MaterialX support significantly. In order to perfect it, I've written an open source @MaterialXcg Fidelity Suite inspired by the @thekhronosgroup / @glTF3D fidelity tool: material-fidelity.ben3d.ca

Captured a vid of sampling a tiny diffusion model with the neural engine (A17 Pro) on my phone. Not very fast but pretty acceptable given everything happened on the iPhone locally

☁️ Tiny diffusion model on @Apple iPhone Neural Engine (ANE) Converted one of my Tiny checkpoints into a CoreML package, and this image was generated in 3.78 seconds with memory use under 140MB on my iPhone 15 Pro. Local, private and energy-efficient 🥂

⛅𝐓𝐢𝐧𝐲 𝐃𝐢𝐟𝐟𝐮𝐬𝐢𝐨𝐧 𝐌𝐨𝐝𝐞𝐥𝐬 I trained before are fairly compact, the left on was from a 𝟗.𝟓 𝐌𝐁 one and the right was from a 𝟐𝟐 𝐌𝐁 one. It is fun to witness how a model learns the statistics which enables generative sampling of that photographic experience.

What I always wanted ! Pure JavaScript OpenEXR I/O Thank you @benhouston3d 🤩

It's been a long time coming but the latest version of three-mesh-bvh brings support for out-of-the-box Line & Point cloud BVH support! Now all your geometries can be fast 🚀 1/3 #threejs #webgl #javascript

Splannequin: Freezing Monocular Mannequin-Challenge Footage with Dual-Detection Splatting TL;DR: How to freeze a dynamic molecular scene when people inevitably move. Contributions: • A Novel Problem Formulation and Benchmark: We are the first to formally address synthesizing high-fidelity, freeze-time videos from monocular MC footage, providing a new benchmark and evaluation protocol. • A Targeted Regularization Framework: We propose a novel method to identify and regularize hidden and defective Gaussians, the primary sources of temporal artifacts, anchoring them to reliable past or future states. • State-of-the-art Performance with Zero Inference Overhead: We improve visual quality and stability in existing methods without architectural changes. As the deformation runs only once for a target instant, we achieve inference speeds exceeding 280 FPS on an RTX 4090.

今年は2次元のtransient renderingを使った作品を提出しました。 運営の皆様お疲れ様でした。 #レイトレ合宿

The same stippling method can create some interesting depth of field effect.

Wow, a *phenomenal* single author paper by @ryichando ryichando.graphics/

My talk from @BetterSoftwareC last week is up on youtube. I present my findings on thread synchronization and job systems that I learned while parallelizing the physics solver. youtube.com/watch?v=Kvsvd67X…

🔥😍🔥

artistic exploration via sampling a tiny neural network's memory. curate your own data and train your own model. #diffusion #neural #generative

Selected more photos from my collection and finished a better dataset for my personal cloudscape model training. At the end, it's the quality of data that matters most for any machine learning solution.

Spent some time going outdoor to shoot more cloud photos. I quickly used around 100 crops and trained a tiny model. Quite happy with the results.

The season finale of Lighter Darker is loads of fun and I am a proud owner of a *Physical* copy of Knoll Light Factory but I really want to try Knoll’s Flare Box

Right-sizing a diffusion model seems important for capturing the different levels of variation of a target distribution. This particular generated sample truly captures what I like about the training set. There are so many dimensions of art making with Machine Learning.

Attention is *not always* our need Removed the Attention modules from the UNet Halved the filter kernel numbers A more compact UNet learned faster and an improved distribution, i.e. better unseen samples.

blossom memories