-
Camera trajectory: You can use the provided trajectories in folder MatrixCitySequence or you own generated trajectory. Please put this folder under the folder
CitySample\Content
. -
Rendering config: You can use the provided config in folder MatrixCityRenderConfig. Please copy these uassets into folder
CitySample\Content\Cinematics
. Ortherwise the engine cannot find these uassets.
The primary aim of the MatrixCity v0.1.0 dataset is to study the core challenges of city-scale neural rendering. To facilitate this, we disabled the directional light, sun dome, fog, and human and car crowds, resulting in a static scene. The environmental settings in our paper are as follows. You can also modify these settings to generate data that meets your requirements.
-
Light:
-
Search
light
inOutliner
, including sky light and directional light. -
In our paper, we set the directional light invisible, ie.
Visible
as False. -
In your work, you can set the directional light visible. After selecting the asset, you can adjust the light intensity, angle, color and so on for both sky light and directional light in the
Details
panel.
-
-
Sun dome:
-
Fog:
-
Human and Car crowds:
-
Search
spawn
inOutliner
, includingBP_MassCrowdSpawner
(number of moving pedestrians),BP_MassTrafficParkedVehichleSpawner
(number of static parked cars), andBP_MassTrafficVehicleSpwaner
(number of moving cars). -
In our paper, we set the
count
ofBP_MassCrowdSpawner
,BP_MassTrafficVehicleSpwaner
andBP_MassTrafficParkedVehichleSpawner
to 0 in theDetails
panel, resulting a static scene. -
In your work, after selecting the asset, you can adjut the count of moving pedestrians, static parked cars and moving cars in the
Details
panel to simulate the real-world dynamic scenes.
-
-
PostProcessVolume:
-
Search
post
inOutliner
, includingPostProcessVolume_Sandbox_only
,PostProcessVolumePostToggle_Sandbox_only
andPostProcessVolume_WP
. -
In our work, we set the
Vignette Intensity
ofPostProcessVolume_Sandbox_only
,PostProcessVolumePostToggle_Sandbox_only
andPostProcessVolume_WP
to 0.0 in theDetails
panel for appearance consistency.Vignette
is an effect that simulates the darkening in real-world camera lenses. High quality lens try to compensate for this effect. The effect is mostly noticeable near the edges of the image. Please refer to vignette for more details. -
In your work, if you want to simulate this camera imperfection, you can adjust this value for the relevant camera in the
Details
panel.
-
4. You can control the rendering pass in the section Exports
. Our default rendering config only enable rgb
, depth
and normal
pass. You can enable other passes you want. You can also control the output format by changing Extension
of each path. Note that output of exr format is in the linear space and output of png format is in the sRGB space, which needs gamma correction for mutual conversion.
5. You can control the output directory, output name format, and output resolution in the section Output
. After finishing setting the render config, remember to click the Accept
icon.
6. (Optional) If you change the config file provided, you can click this icon to create your own config file, which is saved in folder CitySample/Content/Cinematics
.
7. Click the Render(local)
icon to render the images. Note that the first running requires compiling the shaders, which will take a long time. Just wait patiently. Output will be saved in the folder you specify in the render config.
The default camera aspect ratio is 16:9. If your output resolution is not this ratio, you should set Constrain Aspect Ratio
as False in the Details
panel after selecting the cine camera actor. We have changed this setting in our provided camera trajectories of street view. Please note that you need to change this setting for your own generated trajectories, whose output resolution ratio is not 16:9.
1. To achieve the highest possible image quality, we use anti-aliasing during the rendering process. This will make the edges of the depth smooth and brings noises when converting depth into poing cloud. Please refer to Noise in depth gt for more details. We only provide the depth map rendered with anti-alias.
If you want to use the depth map to guide reconstruction, depth map rendered with anti-alias is more suitable because rgb is also rendered with anti-alias. You can download our provided depth map.
If you want to extract geometry from depth map, you should rendered without anti-alias. You just change the
Spatial Sample Count
from8
to1
in rendering config file to disable the anti-alias.