Once the 3D scene is ready, Predict Engine can be started and the scene will automatically be configured from Unreal Engine. The simulation is started from a dedicated window available via the menu "UVR/Engine View" or using the Predict Engine button on the header tab.
Make sure the Predict Engine server has been properly installed on your machine and is correctly referenced in the Preferences before starting a render. See how to Set up Predict Engine for more details.
The view contains the following buttons :
Path tracer selection : Defines whether the path tracer is RGB or Spectral. NB : the level content must be reloaded on the Predict Engine server when the path tracer mode is changed.
Play/Pause/Reload/Stop :
Start the render,
Pause the render,
Force a reload of all the level content on the server,
Stop the render.
Path tracer selection : Save the render to a PNG file, an EXR file with the HDR content of the current view, or an EXR file with all the computed data (radiance, normals, depth,...).
Preferences and documentation :
Open the preferences window (same as menu "UVR/Preferences"),
Open this documentation.
Settings :
Show FPS/Stats inside the viewport,
Adjust the tone mapping settings.
Resolution selection : defines the resolution/ratio of the computed simulation. New resolutions and ratios can be defined in the Preferences.
Hand mode : when this mode is selected, you can move the simulation in the window using the left mouse button. You can also move the simulation in any mode using the mouse scroll button,
Zoom mode : when this mode is selected, you can zoom on the rendered image in the window using the left mouse button. You can also zoom on the rendered image in any mode using the mouse scroll.
Center simulation : resets the zoom and position of the simulation in the window to center it and fill in the available space,
Zoom on pixels : zoom in the simulation to show each pixel RGBA values,
1:1 Scale : resets the zoom and position of the simulation in the window to center it and view it with a 1:1 pixel scale.
Camera Selection and Transform Definition :
Left button : defines which camera in the level is currently being rendered,
Right button : defines whether the position of the selected camera is defined by its Transform component or by the transform of the camera in the Editor view.
Channel selection : defines which channel(s) of the camera are displayed on screen. The available channels depend on the path tracer type.
Transform Definition = Camera View
Transform Definition = Editor View
The field of view of the Editor camera can be changed in the Editor view settings menu :
Tracing light paths from the camera and sorting rays drastically help to reduce rendering times. Nevertheless, the global illumination process remains a computation-intensive task. Hundreds to thousands of light path samples per pixel (spp) might still be necessary to obtain accurate results.
In the case of Path Tracing (and Monte-Carlo methods in general), insufficient light path samples translate to noise in the final image. Such noise can be removed by generating more samples per pixel, thus increasing the rendering time. This can be observed in the following examples (results obtained using a workstation powered by 3 NVIDIA RTX 2080 graphics cards):
32 spp (computed in less than 1 second)
512 spp (computed in 3 seconds)
131 000 spp (computed in 15 minutes)
The correct amount of path samples required to obtain noise-free images depends on the scene content and, more specifically, on the complexity of the paths that light follows before reaching the camera.
For instance, indoor scenes take longer to render when most light comes from the external environment and must cross windows to lit the objects. In the same vein, caustics effects due to refractions or shiny surface reflections might require more time to render smoothly as they involve two or more bounces between the camera and the light source. This can be observed in the following examples:
32 spp (computed in less than 1 second)
512 spp (computed in 3 seconds)
131 000 spp (computed in 15 minutes)
As can be seen, noise remains visible in the caustics areas at 512 spp (middle image). In contrast, at equivalent spp, the result obtained in the previous example with the textile material is almost noise-free.