Serial Founder | AI | Games | Graphics.
SF Bay Area, United States
Joined May 2008
- Tweets1.9K
- Following665
- Followers1.2K
- Likes161
Splatcatcher 1.3 is out!
New: Splatcatcher Forge, a Mac/PC companion that drives LichtFeld Studio, Spirula Studio and more, so you go from capture to trained splat right from your iPhone.
Plus: new native splat viewing via iOS 27 RealityKit. Still free. ❤️
iOS: apps.apple.com/app/id6759800…
Forge (Mac/PC): github.com/warpgatelabs/spla…
Splatcatcher is a free iOS app which accurately captures scenes using iPhone Pro's LiDAR and other sensors, and exports COLMAP for LichtFeld Studio, Brush, Postshot, and wherever else you like to train splats.
The project aims to produce the highest-quality output of any capture app, and this new version makes huge strides in accuracy with pose refinement. The scene above was captured on an iPhone 14 Pro at the Rijksmuseum in Amsterdam during final testing last week.
New Features in 1.2:
- On-device pose refinement: Refine tightens the camera poses using the LiDAR depth and corrects for autofocus drift. Frames it can't align are dropped, and the point cloud is rebuilt from what's left. Pro-tip: Try running this more than once...
- COLMAP export with intrinsics stored for every frame to accurately support autofocus changes
- 4K capture on supported devices, with blur rejection and flying-pixel filtering
- By popular demand: Shutter speed control. If it gets too dark to hold the selected speed, capture switches back to auto
- Drag the live point cloud during capture to check your coverage from another angle
- The display no longer sleeps during long captures
- Devices without LiDAR now show a pleasant error message instead of crashing...
Arash Keshmirian retweeted
Self-driving cars are fun because you never see competing SaaS products having a literal standoff in the street
RSR v0.9.0 is out — the fastest native Windows viewer for 3D Gaussian Splatting just got a major speed boost.
> ~1.8x faster: 80 → 143 FPS on a 6M splat scene at 4K
> Quality presets (Low/Med/High)
> Built on raw D3D12 + NVIDIA DLSS
> OpenXR support for VR
Free Download @ GitHub: github.com/warpgatelabs/RSR/…
Arash Keshmirian retweeted
The strange part of this moment is not that some people suddenly became wealthy. Every major platform shift creates outlier outcomes.
The strange part is that so many ambitious people seem to view the end goal as “make $20M and retire early" if what you are saying is accurate.
Did any generational founder we admire in the valley start the company that took off thinking, “How fast can I make enough money to stop working?” No chance.
Somewhere along the way, our work shifted from craft, invention, mastery, massive competition, and contribution into a spreadsheet calculation about escape velocity.
If the ultimate reward for your career is leaving it, you are probably climbing the wrong mountain.
The people who create enduring companies, technologies, industries, and institutions are rarely motivated primarily by retirement math. They are driven by curiosity, obsession, mission, impact, and the desire to build something excellent. Money is usually a byproduct, not really the point.
A culture optimized around “how quickly can I cash out?” produces mercenaries instead of insanely passionate builders. And mercenary cultures rarely create truly great things.
Arash Keshmirian retweeted
a lovely addon called Gauss Cannon lets you assign an arbitrary mesh (/multiple) and have a camera-per-face.
BYO splatting strat (multiple overlapping spheres / one large dome / whatever serves the use case), just the right amount of control!
Arash Keshmirian retweeted
splat troubleshooting guide:
If you ever export a splat from NerfStudio that's transparent, let it train for another 1000 steps
what's happening is the alphas get reset every 3k steps, so here when i exported at 12k steps, it was right when the alphas reset, hence the transparency.
you can adjust the reset frequency with reset_every: int = 3000
this might've been obvious to some, but I was confused by it last night and found no discussions/explanation about it online. so posting it here to help others
Incredibly high-quality 4DGS capture work!
The (Processing) Level Required to Be Super
We’ve had clients ask us whether our 3D/4DGS capture and processing pipeline holds up in high-end VFX cinematic shots. Last year we were fortunate to work on the Superman production with Framestore (capturing Superman’s parents for a hologram), with our work shown on IMAX screens ~30 m high—so we’re confident the quality delivers.
For this R&D preview, we ran test captures using an affordable off-the-shelf temporal plugin to reduce flicker and improve stability over time.
We recorded @ 48fps from our 196x machine vision cameras and trained the 4DGS using Postshot at Ultra quality ~20 million splat resolution.
What you’re seeing are stop-motion 3DGS frames. Each frame is an individual Gaussian Splat—processed to ~20 million splats for the full scene, then cropped and cleaned to ~6 million splats per frame.
Our system—and the 3DGS process—picks up fine details like stray fine hairs, skin pores and woven cloth threads.
Thanks for watching & stay super!
Learn more @ ir-ltd.net
#3DGS #4DGS #VFX #GaussianSplatting #IMAX #VolumetricCapture #Superman
4K Version here - youtu.be/fdMiypD3VME
Arash Keshmirian retweeted
IBM just released Granite 4.0, their latest series of small language models! These models excel at agentic workflows (tool calling), document analysis, RAG, and more. 🚀
The "Micro" (3.4B) model can even run 100% locally in your browser on WebGPU, powered by 🤗 Transformers.js!
Arash Keshmirian retweeted
An up-close view of TriangleSplatting+.
Still hard to believe that there's a gradient leading to this dataset and that ADAM on GPUs is able to find it. Unthinkable just 5 years ago.
Thanks @taiyasaki for putting the PLY on the project page.
📢📢📢 𝐓𝐫𝐢𝐚𝐧𝐠𝐥𝐞 𝐒𝐩𝐥𝐚𝐭𝐭𝐢𝐧𝐠+: 𝐃𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐭𝐢𝐚𝐛𝐥𝐞 𝐑𝐞𝐧𝐝𝐞𝐫𝐢𝐧𝐠 𝐰𝐢𝐭𝐡 𝐎𝐩𝐚𝐪𝐮𝐞 𝐓𝐫𝐢𝐚𝐧𝐠𝐥𝐞𝐬.
– Project: trianglesplatting2.github.io
– ArXiv: arxiv.org/pdf/2509.25122
– ⚠️Code released on October 8th
Arash Keshmirian retweeted
Instant4D: 4D Gaussian Splatting in Minutes
Contributions:
• We propose INSTANT4D, a modern and fully automated pipeline that reconstructs casual monocular videos within minutes, achieving a 30× speed up.
• We introduce a grid pruning strategy that reduces the number of Gaussians by 92%, preserving occlusion structures and enabling scalability to long video sequences.
• We present a simplified, isotropic, motion-aware 4DGS formulation in a monocular setup, achieving 29% better performance than current state-of-the-art methods on the Dycheck dataset.
I've finally released Gauss Cannon for Blender. It aims to be the go-to solution for exporting Blender scenes to Gaussian Splats using LichtFeld Studio (or Postshot) for processing. Please give it a try and let me know what you think! github.com/keshmirian/gauss-…
With the exodus of Gaussian Splat folks away from PostShot ($$$ now), a new ecosystem emerges around the free Brush and LichtFeld Studio tools. I've been working on a new plugin for Blender called Splat-Tools to do my part.
Test: Rendering my fav @LEGO_Group sets!
[1/2]
[2/2] Additional thanks to @playcanvas for their amazing web viewer and new efficient SOG splat format. I'll be releasing Blender Splat-Tools in a few days.
superspl.at/view?id=aa3ffc8f
Arash Keshmirian retweeted
No, you are probably not running the R1 model locally. The actual Deepseek R1 is 600GB, with the heavily quantized version sitting at 131GB. The "R1 Distilled llama3 or Qwen" youare running is basically Llama3 or Qwen that's trained to behave like R1, but they are not R1.
