Skip to content

Instantly share code, notes, and snippets.

@sadanand1120

sadanand1120/camera-conventions.md

Last active March 7, 2026 14:07
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save sadanand1120/a23d82e785dbe646ea2e250cda44d76f to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save sadanand1120/a23d82e785dbe646ea2e250cda44d76f to your computer and use it in GitHub Desktop.
Download ZIP
Raw
camera-conventions.md

Imagine you (taker) are standing behind camera, i.e., taking a photo using the camera, and a viewer is standing in front of camera.

OpenGL / Blender / NeRF / Nerfstudio

opengl
  • +X right
  • +Y up
  • +Z towards taker aka backwards, so "look-at" direction is -Z

OpenCV / standard CV convention / COLMAP

opencv
  • +X right
  • +Y down
  • +Z towards viewer aka forwards, so "look-at" direction is +Z

PyTorch3D

pytorch3d
  • +X left
  • +Y up
  • +Z towards viewer aka forwards, so "look-at" direction is +Z

Quick conversions

Assume you have (N, 3) numpy array pointcloud.

  • OpenCV to/from OpenGL: new_pts = pts * np.array([1.0, -1.0, -1.0]
  • OpenGL to/from PyTorch3D: new_pts = pts * np.array([-1.0, 1.0, -1.0]
  • OpenCV to/from PyTorch3D: new_pts = pts * np.array([-1.0, -1.0, 1.0]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment