代码拉取完成，页面将自动刷新
At the moment, we provide bindings for essential and fundamental matrix estimation as well as absolute pose estimation.
Clone the repository and its submodules by running:
git clone --recursive git@github.com:mihaidusmanu/pycolmap.git
Preferred: pycolmap is now available for Python 3.7 and Python 3.8 via pypi:
pip install pycolmap
These wheels were built using John Lambert's wheel-builder.
If you instead prefer to build the wheels from source, COLMAP should be installed as a library before proceeding. Please refer to the official website for installation instructions. PyCOLMAP can then be installed using pip:
pip install ./
To install PyCOLMAP on Windows, we recommend to install COLMAP with vcpkg. From your vcpkg directory, run
.\vcpkg.exe install colmap --triplet=x64-windows --head
Then set the CMAKE_TOOLCHAIN_FILE environment variable to your vcpkg\scripts\buildsystems\vcpkg.cmake path.
Example (powershell):
$env:CMAKE_TOOLCHAIN_FILE='C:\Workspace\vcpkg\scripts\buildsystems\vcpkg.cmake'
Finally go to the PyCOLMAP folder and run
py -m pip install ./
PyCOLMAP build is generally tested with the latest commit of the COLMAP dev branch.
Using a previous COLMAP build might not work.
The current bindings are compatible with numpy arrays for both 2D and 3D points. The camera parameters should be sent as a Python dictionary with the following template:
{
'model': COLMAP_CAMERA_MODEL_NAME,
'width': IMAGE_WIDTH,
'height': IMAGE_HEIGHT,
'params': EXTRA_CAMERA_PARAMETERS_LIST
}
Please refer to colmap - src/base/camera_models.h for more details regarding camera models and parameters.
For instance, the following snippet runs absolute pose estimation for a pinhole camera given 2D-3D correspondences:
import pycolmap
# Parameters:
# - points2D: Nx2 array; pixel coordinates
# - points3D: Nx3 array; world coordinates
# - camera_dict: dictionary
# Named parameters
# - max_error_px: float; RANSAC inlier threshold in pixels
answer = pycolmap.absolute_pose_estimation(
points2D, points3D,
{
'model': 'SIMPLE_PINHOLE',
'width': width,
'height': height,
'params': [focal_length, cx, cy]
}
)
# Returns:
# - dictionary containing the RANSAC output
import pycolmap
# Parameters:
# - tvec: List of 3 floats, translation component of the pose (world to camera)
# - qvec: List of 4 floats, quaternion component of the pose (world to camera)
# - points2D: Nx2 array; pixel coordinates
# - points3D: Nx3 array; world coordinates
# - inlier_mask: array of N bool; true -> corresponding value in points2D/points3D is an inlier
# - camera_dict: dictionary
answer = pycolmap.pose_refinement(
tvec, qvec, points2D, points3D, inlier_mask,
{
'model': 'SIMPLE_PINHOLE',
'width': width,
'height': height,
'params': [focal_length, cx, cy]
}
)
# Returns:
# - dictionary containing the RANSAC output
import numpy as np
import pycolmap
from PIL import Image, ImageOps
# Input should be grayscale image with range [0, 1].
with open('image.jpg', 'rb') as f:
img = Image.open(f)
img = img.convert('RGB')
img = ImageOps.grayscale(img)
img = np.array(img).astype(np.float) / 255.
# Parameters:
# - image: HxW float array
# Named parameters:
# - num_octaves: int (4)
# - octave_resolution: int (3)
# - first_octave: int (0)
# - edge_thresh: float (10)
# - peak_thresh: float (0.01)
# - upright: bool (False)
keypoints, scores, descriptors = pycolmap.extract_sift(img)
# Returns:
# - keypoints: Nx4 array; format: x (j), y (i), sigma, angle
# - scores: N array; DoG scores
# - descriptors: Nx128 array; L2-normalized descriptors
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手