pyrender基于PyRender的深度图渲染

Posted 2023-03-24

一、安装pyrender

• 安装pyrender: https://pyrender.readthedocs.io/en/latest/install/index.html
• 跑通Offscreen Rendering程序，验证环境是否配置成功：https://pyrender.readthedocs.io/en/latest/examples/quickstart.html

二、加载Mesh模型与相机参数

• 加载Mesh模型
  • colmap fused.ply --> PoissonRecon
• 相机参数的转换
  • colmap (world-to-camera) vs pyrender (camera-to-world)
  • y-z axis reverse

# Creates the scene and adds the mesh.
scene = pyrender.Scene()
scene.add(mesh)

# Creates a renderer of suitable height and width
cam_data = cameras[images[i].camera_id]
renderer = pyrender.OffscreenRenderer(cam_data.width,
                                      cam_data.height)

# Extracts focal length and principal point information.
# Important: This code snippet assumes that the camera
# is a PINHOLE camera.
fx = np.float32(cam_data.params[0])
fy = np.float32(cam_data.params[1])
cx = np.float32(cam_data.params[2])
cy = np.float32(cam_data.params[3])

# Extracts the pose of the image.
# Note that pyrender poses transform from camera
# to world coordinates while colmap poses transform
# from world to camera.
R = np.asmatrix(qvec2rotmat(images[i].qvec)).transpose()
T = np.identity(4)
T[0:3,0:3] = R
T[0:3,3] = -R.dot(images[i].tvec)
# Takes into account that colmap uses the computer vision
# camera coordinate system (x = right, y = down, z = front)
# while pyrender uses the computer graphics conventions
# (x = right, y = up, -z = front).
T[:, 1:3] *= -1

# Sets up the camera with the intrinsics and extrinsics.
pyrender_camera = pyrender.IntrinsicsCamera(fx, fy, cx, cy,
                                            zfar=800.0)  # 425~935
cam_node = scene.add(pyrender_camera, pose=T)

参考资料：

https://github.com/colmap/colmap/issues/704#issuecomment-954161261