File size: 7,238 Bytes
8c48cce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
import os
import json
import numpy as np

from utils.math_utils import world_to_view, projection_matrix

# Y down, Z forward
def load_camera(camera_info):
    """Load camera parameters from camera info dictionary"""
    # Extract camera parameters
    camera_id = camera_info["camera_id"]
    camera_to_world = np.asarray(camera_info["camera_to_world"], dtype=np.float64)
    
    # Change from OpenGL/Blender camera axes (Y up, Z back) to COLMAP (Y down, Z forward)
    camera_to_world[:3, 1:3] *= -1
    
    # Calculate world to camera transform
    world_to_camera = np.linalg.inv(camera_to_world).astype(np.float32)
    
    
    # Extract rotation and translation
    R = world_to_camera[:3, :3]
    T = world_to_camera[:3, 3]
    
    
    world_to_camera[3, 3] = 1.
    world_to_camera = world_to_camera.T

    
    width = camera_info.get("width")
    height = camera_info.get("height")
    fx = camera_info.get("focal")
    fy = camera_info.get("focal")
    cx = width / 2
    cy = height / 2
    
    # Calculate field of view from focal length
    fovx = 2 * np.arctan(width / (2 * fx))
    fovy = 2 * np.arctan(height / (2 * fy))
    
    # Create view matrix
    view_matrix = world_to_view(R=R, t=T)
    
    # Create projection matrix
    znear = 0.01
    zfar = 100.0
    proj_matrix = projection_matrix(fovx=fovx, fovy=fovy, znear=znear, zfar=zfar).T
    full_proj_matrix = world_to_camera @ proj_matrix
    
    # Calculate other parameters
    tan_fovx = np.tan(fovx * 0.5)
    tan_fovy = np.tan(fovy * 0.5)
    
    camera_center = np.linalg.inv(world_to_camera)[3, :3]
    
    # Handle camera type and distortion
    camera_model = camera_info.get("camera_model", "OPENCV")
    if camera_model == "OPENCV" or camera_model is None:
        camera_type = 0  # PERSPECTIVE
    elif camera_model == "OPENCV_FISHEYE":
        camera_type = 1  # FISHEYE
    else:
        raise ValueError(f"Unsupported camera_model '{camera_model}'")
    
    # Get distortion parameters
    distortion_params = []
    for param_name in ["k1", "k2", "p1", "p2", "k3", "k4"]:
        distortion_params.append(camera_info.get(param_name, 0.0))
    
    camera_params = {
        'R': R,
        'T': T,
        'camera_center': camera_center,
        'view_matrix': view_matrix,
        'proj_matrix': proj_matrix,
        'full_proj_matrix': full_proj_matrix,
        'tan_fovx': tan_fovx,
        'tan_fovy': tan_fovy,
        'fx': fx,
        'fy': fy,
        'cx': cx,
        'cy': cy,
        'width': width,
        'height': height,
        'camera_to_world': camera_to_world,
        'world_to_camera': world_to_camera,
        'camera_type': camera_type,
        'distortion_params': np.array(distortion_params, dtype=np.float32)
    }
    
    return camera_params

def load_camera_from_json(input_path, camera_id=0):
    """Load camera parameters from camera.json file"""
    camera_file = os.path.join(os.path.dirname(input_path), "cameras.json")
    if not os.path.exists(camera_file):
        print(f"Warning: No cameras.json found in {os.path.dirname(input_path)}, using default camera")
        return None
    
    try:
        with open(camera_file, 'r') as f:
            cameras = json.load(f)
        
        # Find camera with specified ID, or use the first one
        camera = next((cam for cam in cameras if cam["id"] == camera_id), cameras[0])
        
        # Use load_camera to process the camera parameters
        return load_camera(camera)
        
    except Exception as e:
        print(f"Error loading camera from cameras.json: {e}")
        return None

def load_camera_colmap(cam_info):
    """
    Load camera from COLMAP format (dust3r output) with exact compatibility to original load_camera.
    
    Args:
        cam_info: Dictionary containing:
            - width, height: image dimensions
            - fx, fy: focal lengths
            - cx, cy: principal point
            - camera_id: unique identifier
            - R: rotation matrix (world-to-camera rotation)
            - T: translation vector (world-to-camera translation)
            - Optional: camera_model, distortion params
    """
    # Extract camera parameters
    camera_id = cam_info["camera_id"]
    
    # Use provided R and T directly (COLMAP convention - world to camera)
    R = cam_info['R'] 
    T = cam_info['T']  # This is world-to-camera translation
    
    # Build world-to-camera matrix
    world_to_camera = np.eye(4, dtype=np.float64)
    world_to_camera[:3, :3] = R
    world_to_camera[:3, 3] = T
    
    # Invert to get camera-to-world
    camera_to_world = np.linalg.inv(world_to_camera).astype(np.float64)
    
    # IMPORTANT FIX: Ensure Z direction is correctly oriented for COLMAP convention
    # COLMAP uses +Z forward, so no need to flip Z axis
    # If frustums are still backwards, uncomment this line:
    # camera_to_world[:3, 2] *= -1  # Flip Z axis if needed
    
    # Recalculate world_to_camera after any modifications
    world_to_camera = np.linalg.inv(camera_to_world).astype(np.float32)
    
    # Extract intrinsics
    width = cam_info.get("width")
    height = cam_info.get("height")
    fx = cam_info.get("fx", cam_info.get("focal", width * 0.7))
    fy = cam_info.get("fy", cam_info.get("focal", height * 0.7))
    cx = cam_info.get("cx", width / 2)
    cy = cam_info.get("cy", height / 2)
    
    # Calculate field of view from focal length
    fovx = 2 * np.arctan(width / (2 * fx))
    fovy = 2 * np.arctan(height / (2 * fy))
    
    # Create view matrix using the original R and T
    view_matrix = world_to_view(R=R, t=T)
    
    # Create projection matrix
    znear = 0.01
    zfar = 100.0
    proj_matrix = projection_matrix(fovx=fovx, fovy=fovy, znear=znear, zfar=zfar).T
    full_proj_matrix = world_to_camera @ proj_matrix
    
    # Calculate other parameters
    tan_fovx = np.tan(fovx * 0.5)
    tan_fovy = np.tan(fovy * 0.5)
    
    # IMPORTANT FIX: Correctly calculate camera center
    camera_center = camera_to_world[:3, 3]  # Extract translation from c2w matrix
    
    # Handle camera type and distortion
    camera_model = cam_info.get("camera_model", "OPENCV")
    if camera_model == "OPENCV" or camera_model is None:
        camera_type = 0  # PERSPECTIVE
    elif camera_model == "OPENCV_FISHEYE":
        camera_type = 1  # FISHEYE
    else:
        camera_type = 0  # Default to PERSPECTIVE
    
    # Get distortion parameters
    distortion_params = []
    for param_name in ["k1", "k2", "p1", "p2", "k3", "k4"]:
        distortion_params.append(cam_info.get(param_name, 0.0))
    
    # Return camera parameters
    camera_params = {
        'R': R,
        'T': T,
        'camera_center': camera_center,
        'view_matrix': view_matrix,
        'proj_matrix': proj_matrix,
        'full_proj_matrix': full_proj_matrix,
        'tan_fovx': tan_fovx,
        'tan_fovy': tan_fovy,
        'fx': fx,
        'fy': fy,
        'cx': cx,
        'cy': cy,
        'width': width,
        'height': height,
        'camera_to_world': camera_to_world,
        'world_to_camera': world_to_camera,
        'camera_type': camera_type,
        'distortion_params': np.array(distortion_params, dtype=np.float32)
    }
    
    return camera_params