An NVIDIA research project for training neural fields, with a NeRF mode, source code, and a viewer.
What Instant NGP trains
In NeRF mode, the NVIDIA project builds a field that describes the color and density of the scene for different viewing directions. Entry includes photos and camera position for each image. The difference between Instant NGP work is related to the way to encode coordinates and speed up calculations on a graphics processor. The repository includes other tasks, such as presenting an image to the network. NeRF mode is required to test the three-dimensional scene.
Check a novel viewpoint
Start with an authorized set of photographs of a wooden chair. If you shoot your own, walk around it and keep adjacent images overlapping. After estimating the camera poses, check that the cameras are arranged around the chair in the expected order. Train a small scene and change the viewpoint in the viewer. Inspect the gap between the legs and the thin edge of the seat. Camera pose errors often show up at these boundaries.
Write down the parameters of the first attempt. When you repeat, you can change one parameter and understand what happened to the detail and memory consumption.
Make the run reproducible
The official repository describes the build, CUDA requirements and data formats. On another machine, the speed and available scene sizes may vary. Save the code version along with settings and source angles. To transfer a colleague, specify in advance whether an interactive view or a ready render is needed. One option is easier to view, the second is easier to embed in the video.
Suitable tasks
- Teach your subject through a series of angles
- Compare visual quality with two memory settings
Strengths
- Research code with interactive viewing
- Support for multiple classes of neural representations
Limitations
- Requires NVIDIA CUDA environment
- The result depends on the correctness of the position of the cameras
Selection and conditions
- Free access
- Yes
- Trial period
- Not confirmed
- No sign-up required
- Not confirmed
- Russian interface
- Not confirmed
- Local installation
- Yes
- API
- Not confirmed
“Not confirmed” means there is not enough information. The feature may still be available; check the developer’s terms.
- Input formats
- Image
- Result
- 3D scene
Terms
Local assembly requires a compatible NVIDIA graphics card and environment from the repository. Supported startup options are described in the instructions.From task to first result
Practical example
- Task
- Teach your subject through a series of angles
- What to prepare
- His consecutive photos of one static scene, shot with overlapping angles.
Task brief or prompt
Restore the scene from photos and view it from a new perspective. Leave some of the original species for comparison.
Expected result
A scene that can be viewed in the viewer.
How to check the result
- Compare a few angles with the control images.
- Mark gaps, transparent areas and edges outside the shooting area.
An editorial teaching example of a task brief, not the result of hands-on testing of the service. Check that the required mode is available, that you have rights to the source material and that the intended use of the result is permitted.
Check details
Official sources
This description is based on the developer’s material. Features, prices and availability may change; check the primary source before use.
