Stereo and Multi-Rig Setups¶

This tutorial shows how to render stereoscopic (left/right eye) or multi-camera rig sequences using the camera_offset parameter. Stereo rendering is useful for downstream sensor emulation that requires depth perception (e.g., binocular vision models, disparity estimation).

Understanding Camera Offsets¶

The camera_offset moves the camera in its local coordinate frame. For more details on the camera coordinate system see the coordinate conventions section. In this frame:

X : left (negative) / right (positive)
Y : down (negative) / up (positive)
Z : backward (negative) / forward (positive)

For a typical stereo pair, a human interpupillary distance (IPD) of around 6.5 cm is standard. Each eye is offset by half the IPD along the local X axis:

Left eye:  camera_offset = (-0.0325,  0.0, 0.0)
Right eye: camera_offset = ( 0.0325,  0.0, 0.0)

Using the Python API¶

The offset_camera method offsets the camera in its local coordinate frame for the current frame. After offsetting, call set_camera_keyframe to bake the new position into the animation:

# Left eye: offset 3.25 cm to the left
client.offset_camera((-0.0325, 0.0, 0.0))
client.set_camera_keyframe(frame)

# Right eye: offset 3.25 cm to the right
client.offset_camera((0.0325, 0.0, 0.0))
client.set_camera_keyframe(frame)

Using the CLI¶

The --camera-offset flag applies the offset to every rendered frame automatically. Render each eye view into a separate output directory:

# Render left eye view
visionsim render-animation scene.blend ./left_eye \\
    --camera-offset -0.0325 0 0

# Render right eye view
visionsim render-animation scene.blend ./right_eye \\
    --camera-offset 0.0325 0 0

Note

The offset is applied in local camera coordinates, so it works correctly regardless of the camera’s world-space orientation. The left/right direction always follows the camera’s own +X axis.

Multi-Rig Setups¶

For multi-camera rigs with more than two cameras (e.g., a 3-camera array), simply run render-animation once per camera position with the appropriate offset:

# Three-camera rig at 6.5 cm spacing
visionsim render-animation scene.blend ./cam_left   --camera-offset -0.065 0 0
visionsim render-animation scene.blend ./cam_center --camera-offset  0.0   0 0
visionsim render-animation scene.blend ./cam_right  --camera-offset  0.065 0 0

Combining Stereo Views¶

After rendering both eyes, you can use the Dataset API to inspect or merge the output directories:

from visionsim.dataset import Dataset

left  = Dataset.from_path("./left_eye/frames")
right = Dataset.from_path("./right_eye/frames")

# Both datasets contain the same number of frames with corresponding indices
for (left_img, _), (right_img, _) in zip(left, right):
    # left_img and right_img have shape (H, W, C)
    ...