Camera Lit Capture Parameters

Here we discuss the parameters available for controlling the lit capture of a camera in LychSim. Qualitative examples of the effects are available at the bottom of this page.

Warmup Parameters

Capturing the lit image of a camera in LychSim requires the camera to be warmed up for a few frames before a clean image can be read back. This is a consequence of how UE5 renders the scene:

• Lumen (used for both dynamic global illumination and reflections in ULitCamSensor) is a temporal algorithm — its radiance caches, screen-space probes, and reflection data are built up over multiple frames and are not fully converged on the first render.

• Temporal anti-aliasing (TAA/TSR) and other screen-space effects (SSR, SSAO, bloom history, auto-exposure) likewise accumulate across frames.

• The lit sensor runs with bCaptureEveryFrame = false and bAlwaysPersistRenderingState = true, so these caches only advance when a capture is explicitly requested. Without a few warmup captures, the first readback contains noisy GI, aliased edges, and an unconverged exposure.

There are two ways to warm up the camera:

sim.warmup_cam(cam_id=0, num_steps=20)
img = sim.get_cam_lit(cam_id=0)

img = sim.get_cam_lit(cam_id=0, warmup=20)

The results are similar, but the second approach is significantly faster. The first form invokes the full GetLit pipeline once per warmup step — each step performs a CaptureScene, a GPU -> CPU ReadPixels, an sRGB <-> linear conversion pass, image serialization, and a full network round-trip back to Python. The second form pushes the warmup count down to the sensor, which loops CaptureScene(); FlushRenderingCommands(); internally and only does the expensive readback and round-trip on the final frame, eliminating N-1 readbacks and N-1 network hops.

Average capture times for a 1280x720 lit image using different parameters on an RTX 4090 are as follows:

# Warmup Steps	Warmup Method 1	Warmup Method 2	Warmup Method 2 + Experimental Flag
0	0.36	0.36	0.36
10	1.14	0.44	0.44
20	1.92	0.52	0.52
50	4.28	0.79	0.79
100	8.19	1.23	1.24

Experimental Flag

An experimental flag is also available for get_cam_lit, which toggles a set of post-process and show-flag overrides on the lit sensor intended to produce sharper, more deterministic captures for synthetic-data use cases. Specifically, when the flag is set, ULitCamSensor::CaptureLit disables:

• Temporal anti-aliasing (ShowFlags.TemporalAA = false) — removes temporal accumulation so each captured frame stands on its own, avoiding ghosting between captures of moving objects or cameras.

• Motion blur (MotionBlurAmount = 0, ShowFlags.MotionBlur = false) — avoids velocity-based blur when objects or the camera have moved between frames.

• Depth of field (DepthOfFieldScale = 0, ShowFlags.DepthOfField = false) — forces a fully sharp image regardless of the active post-process volume’s DoF settings.

These effects are useful for photorealism but harmful for most computer-vision data generation (segmentation masks, depth supervision, 6D pose), which is why they are opt-out behind the experimental flag.

Qualitative Comparison

Comparison of lit captures with (1) left: warmup method 1, (2) middle: warmup method 2, and (3) right warmup method 2 + experimental flag. From top to bottom, the number of warmup steps is 0, 10, 20, 50, and 100.

Comparison of lit captures with (1) left: warmup method 1, (2) middle: warmup method 2, and (3) right warmup method 2 + experimental flag. From top to bottom, the number of warmup steps is 0, 10, 20, 50, and 100.