Estimate the scale and shift of the depth map based on the scene sparse point cloud, ussing RANSAC to find the best fit: depth_map_scaled = scale * depth_map + shift Args: scene (dict): The MVS scene data. image_idx (int): The index of the image in the scene. depth_map (numpy.
(scene, image_idx, depth_map, verbose=False)
| 64 | |
| 65 | |
| 66 | def scale_depth_map(scene, image_idx, depth_map, verbose=False): |
| 67 | """ |
| 68 | Estimate the scale and shift of the depth map based on the scene sparse point cloud, |
| 69 | ussing RANSAC to find the best fit: |
| 70 | depth_map_scaled = scale * depth_map + shift |
| 71 | Args: |
| 72 | scene (dict): The MVS scene data. |
| 73 | image_idx (int): The index of the image in the scene. |
| 74 | depth_map (numpy.ndarray): The depth map to be scaled corresponding to the image. |
| 75 | verbose (bool): If True, print debug information. |
| 76 | Returns: |
| 77 | tuple: Scale and shift values. |
| 78 | """ |
| 79 | from sklearn.linear_model import RANSACRegressor |
| 80 | |
| 81 | # Collect 3D points and corresponding depth values |
| 82 | image = scene["images"][image_idx] |
| 83 | image_width = scene["platforms"][image["platform_id"]]["cameras"][image["camera_id"]]["width"] |
| 84 | image_height = scene["platforms"][image["platform_id"]]["cameras"][image["camera_id"]]["height"] |
| 85 | K = np.array(scene["platforms"][image["platform_id"]]["cameras"][image["camera_id"]]["K"]) |
| 86 | R = np.array(scene["platforms"][image["platform_id"]]["poses"][image["pose_id"]]["R"]) |
| 87 | C = np.array(scene["platforms"][image["platform_id"]]["poses"][image["pose_id"]]["C"]) |
| 88 | K = scale_K(K, depth_map.shape[1] / image_width, depth_map.shape[0] / image_height) |
| 89 | depths_sfm = [] |
| 90 | depths_dmap = [] |
| 91 | mean_depth = 0 |
| 92 | for vertex in scene['vertices']: |
| 93 | for view in vertex['views']: |
| 94 | if view['image_id'] == image_idx: |
| 95 | # Project the 3D point to the image plane |
| 96 | # and get the corresponding depth value |
| 97 | Xcam = R @ (vertex['X'] - C) |
| 98 | depth_sfm = float(Xcam[2]) |
| 99 | if depth_sfm <= 0: |
| 100 | break |
| 101 | x = K @ Xcam |
| 102 | x = np.array([x[0]/x[2], x[1]/x[2]]) |
| 103 | depth_dmap = sample_depth_map(depth_map, x) |
| 104 | if depth_dmap <= 0: |
| 105 | break |
| 106 | depths_sfm.append(depth_sfm) |
| 107 | depths_dmap.append(depth_dmap) |
| 108 | mean_depth += depth_sfm |
| 109 | break |
| 110 | if len(depths_sfm) < 2: |
| 111 | return 1.0, 0.0 |
| 112 | mean_depth /= len(depths_sfm) |
| 113 | depths_sfm = np.array(depths_sfm).reshape(-1, 1) |
| 114 | depths_dmap = np.array(depths_dmap).reshape(-1, 1) |
| 115 | |
| 116 | # Define the estimator, with all the functions required by RANSAC |
| 117 | class Estimator: |
| 118 | def __init__(self, scale=1.0, shift=0.0): |
| 119 | self.scale = scale |
| 120 | self.shift = shift |
| 121 | |
| 122 | def fit(self, X, y): |
| 123 | # Solve for scale and shift |
no test coverage detected