Sign In

Forrest Briggs

Practical immersive volumetric video.

Forrest Briggs is a highly skilled individual with a diverse background in machine learning, computer vision, mathematical optimization, and more.

He possesses a strong command of various programming languages like C++, Java, Python, as well as experience with SQL, Hive, Scala, and more.

Forrest has a broad range of experience in developing applications for different platforms including Mac, Windows, Linux, iOS, Android, and Web.

His educational background includes a B.S. in Computer Science from Harvey Mudd College and a Doctor of Philosophy (Ph.D.) in Computer Science from Oregon State University.

Throughout his career, Forrest has held key positions at reputable organizations such as Lifecast Incorporated, Google X, Lyft Level 5, Facebook, Rocket Fuel Inc., and more.

His experience spans from being a Chief Executive Officer to roles like Machine Learning Scientist, Software Engineer, Technical Lead, and Developer across various companies.

Forrest has also interned at JPL (NASA's Jet Propulsion Laboratory) and has worked on projects involving virtual reality, games, advertising, cryptocurrencies, and ecology.

In addition to his technical expertise, he is proficient in using tools and frameworks like OpenGL, OpenCV, TensorFlow, PyTorch, and has skills in using version control systems like git, mercurial, and svn.

Forrest Briggs is a talented individual with a rich professional background and a passion for innovation and technology.

Highlights

Jun 7 · linkedin

Volumetric Video Editor (VVE) 2.5 - 10x faster on Mac, exports WebXR for Vision Pro https://lnkd.in/gfm6vuve

VVE converts VR180 to our volumetric/6DOF format 'ldi3' which can be played in realtime on Vision Pro and Quest in WebXR. This update makes VVE render~10x faster on Mac and now the WebXR templates it generates support Vision Pro.

Version 2.5 Notes

  • Rendering VR180 to ldi3 on Mac is now much faster (about 10x on my Macbook Pro M3). The stereo fisheye depth estimation neural net now runs on Metal instead of the CPU (on Windows it uses CUDA).
  • When generating a WebXR template page, it now uses the latest version of Lifecast's player, which is hosted on a CDN and automatically updated (it no longer requires us to update VVE to update the player). The web template has been simplified to use only one html file instead of multiple files and resources; this makes it simpler to deploy or integrate into your own web projects.
  • The WebXR player now supports Apple Vision Pro.
  • VVE now encodes h265 video output with the hvc1 flag for compatibility with Vision Pro (and still works on Quest).
  • The interface for launching from VVE to a local web page has been simplified, and now requires only a single click instead of several. Also VVE now runs its own web server rather than spawning a separate terminal with a python web server.
  • Exporting to OBJ now generates a multi-layered mesh with alpha channels in a texture atlas, rather than just a single layer. This output is compatible with blender (similar workflow to exporting from Volurama, see https://lnkd.in/gHfUUEuS)

#ai #cg #3d #vfx #vr180 #6dof #avp #quest #visionpro #spatialvideo

Jun 7 · linkedin

lifecast.ai update - Our open source neural radiance field (NeRF) video engine now trains faster, down from 347sec to 209sec per frame on an RTX4090. This is thanks to the work of Matthew Orlando, one of the most expert programmers I know. At the core of our NeRF engine is a neural multi-resolution hashmap, introduced in "Instant Neural Graphics Primitives with a Multiresolution Hash Encoding" by Muller et al (NVidia). We implemented our own version of this in pure LibTorch C++, which is useful because it runs on Apple silicon with Metal acceleration. NVidia's tiny-cuda-nn library includes a super optimized CUDA version, and we are now using that on Windows and Linux for a ~1.6x speedup to NeRF training.

Here's the tricky part- there are examples from NVidia of using tiny-cuda-nn in python with pytorch, but we want to use it in C++ with LibTorch, and there are no examples of that. It's not trivial; it involves managing memory on the GPU with different interfaces between CUDA and LibTorch tensors. Matt implemented the LibTorch C++ interface to make that possible, and we made it open source because it is a basic useful component for anyone trying to implement a state of the art NeRF in C++.

Open source NeRF video engine: https://lnkd.in/gtnDRAC6

Our paper on compressing NeRF video for realtime playback and our NeRF training algorithms: https://lnkd.in/gw2VKGwW

WebXR NeRF video demos for Apple Vision Pro and Quest: https://lifecast.ai/

tiny-cuda-nn integration in C++ update: https://lnkd.in/gdPXx-ei

tiny-cuda-nn library from NVida: https://lnkd.in/g3p-Dx_c

Volurama - our Mac and Windows GUI that runs the same NeRF engine https://volurama.com

Forrest Briggs
Add to my network

Location

San Mateo, California, United States