Editor's note: This post is part of the AI Decoded series, which demystifies AI by making the technology more accessible, and which showcases new hardware, software, tools and accelerations for RTX PC users.Imagine a gorgeous vista, like a cliff along the water's edge. Even as a 2D image, the scene would be beautiful and inviting. Now imagine exploring that same view in 3D - without needing to be there.
NVIDIA RTX technology-powered AI helps turn such an imagination into reality. Using Instant NeRF, creatives are transforming collections of still images into digital 3D scenes in just seconds.
Simply Radiant A NeRF, or neural radiance field, is an AI model that takes 2D images representing a scene as input and interpolates between them to render a complete 3D scene. The model operates as a neural network - a model that replicates how the brain is organized and is often used for tasks that require pattern recognition.
Using spatial location and volumetric rendering, a NeRF uses the camera pose from the images to render a 3D iteration of the scene. Traditionally, these models are computationally intensive, requiring large amounts of rendering power and time.
A recent NVIDIA AI research project changed that.
Instant NeRF takes NeRFs to the next level, using AI-accelerated inverse rendering to approximate how light behaves in the real world. It enables researchers to construct a 3D scene from 2D images taken at different angles. Scenes can now be generated in seconds, and the longer the NeRF model is trained, the more detailed the resulting 3D renders.
NVIDIA researchers released four neural graphics primitives, or pretrained datasets, as part of the Instant-NGP training toolset at the SIGGRAPH computer graphics conference in 2022. The tools let anyone create NeRFs with their own data. The researchers won a best paper award for the work, and TIME Magazine named Instant NeRF a best invention of 2022.
In addition to speeding rendering for NeRFs, Instant NeRF makes the entire image reconstruction process accessible using NVIDIA RTX and GeForce RTX desktop and laptop GPUs. While the time it takes to render a scene depends on factors like dataset size and the mix of image and video source content, the AI training doesn't require server-grade or cloud-based hardware.
NVIDIA RTX workstations and GeForce RTX PCs are ideally suited to meet the computational demands of rendering NeRFs. NVIDIA RTX and GeForce RTX GPUs feature Tensor Cores, dedicated AI hardware accelerators that provide the horsepower to run generative AI locally.
Ready, Set, Go Get started with Instant NeRF to learn about radiance fields and experience imagery in a new way.
Developers and tech enthusiasts can download the source-code base to compile. Nontechnical users can download the Windows installers for Instant-NGP software, available on GitHub.
While the installer is available for a wide range of RTX GPUs, the program performs best on the latest-architecture GeForce RTX 40 Series and NVIDIA RTX Ada Generation GPUs.
The Getting Started With Instant NeRF guide walks users through the process, including loading one of the primitives, such as NeRF Fox, to get a sense of what's possible. Detailed instructions and video walkthroughs - like the one above - demonstrate how to create NeRFs with custom data, including tips for capturing good input imagery and compiling codebases (if built from source). The guide also covers using the Instant NeRF graphical user interface, optimizing scene parameters and creating an animation from the scene.
The NeRF community also offers many tips and tricks to help users get started. For example, check out the livestream below and this technical blog post.
Show and Tell Digital artists are composing beautiful scenes and telling fresh stories with NVIDIA Instant NeRF. The Instant NeRF gallery showcases some of the most innovative and thought-provoking examples, viewable as video clips in any web browser.
Here are a few:
Through the Looking Glass by Karen X. Cheng and James Perlman A pianist practices her song, part of her daily routine, though there's nothing mundane about what happens next. The viewer peers into the mirror, a virtual world that can be observed but not traversed; it's unreachable by normal means. Then, crossing the threshold, it's revealed that this mirror is in fact a window into an inverted reality that can be explored from within. Which one is real?
https://blogs.nvidia.com/wp-content/uploads/2024/04/studio-ai-decoded-karenx-v9-1280x680-1.mp4
Meditation by Franc Lucent As soon as they walked into one of many rooms in Nico Santucci's estate, Lucent knew they needed to turn it into a NeRF. Playing with the dynamic range and reflections in the pond, it presented the artist with an unknown exploration. They were pleased with the softness of the light and the way the NeRF elevates the room into what looks like something out of a dream - the perfect place to meditate. A NeRF can freeze a moment in a way that's more immersive than a photo or video.
https://blogs.nvidia.com/wp-content/uploads/2024/04/studio-ai-decoded-franc-lucent-meditation-1280x680-1.mp4
Zeus by Hugues Bruy re These rendered 3D scenes with Instant NeRF use the data Bruy re previously captured for traditional photogrammetry using mirrorless digital cameras, smartphones, 360-degree cameras and drones. Instant NeRF gives him a powerful tool to help preserve and share cultural artifacts through online libraries, museums, virtual-reality experiences and heritage-conservation projects. This NeRF was trained using a dataset of photos taken with an iPhone at the Royal Ontario Museum.
https://blogs.nvidia.com/wp-content/uploads/2024/04/studio-ai-decoded-hugues-zeus-1280x680-1.mp4
From Image to Video to Reality Transforming images into a 3D scene with AI is cool. Stepp
