common: add Python utility for reading MVS interface archives

scripts/python/MvsReadDMAP.py

@@ -1,5 +1,5 @@
 '''
-Example usage of MvsUtils.py for exporting dmap file content.
+Example usage of MvsUtils.py for reading DMAP file content.
 
 usage: MvsReadDMAP.py [-h] [--input INPUT] [--output OUTPUT]
 '''
@@ -25,9 +25,9 @@ def exportDMAPContent(dmap_path):
 
 def main():
   parser = ArgumentParser()
-  parser.add_argument('-i', '--input', type=str, required=True, help='path to the depth map directory')
-  parser.add_argument('-t', '--threads', type=int, default=int(os.cpu_count() * 0.5) - 1, help='number of parallel computations')
-  parser.add_argument('-o', '--output', type=str, required=True, help='path to the output directory')
+  parser.add_argument('-i', '--input', type=str, required=True, help='Path to the DMAP file directory')
+  parser.add_argument('-t', '--threads', type=int, default=int(os.cpu_count() * 0.5) - 1, help='Number of parallel computations')
+  parser.add_argument('-o', '--output', type=str, required=True, help='Path to the output directory')
   args = parser.parse_args()
 
   dmap_paths = glob(os.path.join(args.input, '*.dmap'))

scripts/python/MvsReadMVS.py

@@ -0,0 +1,35 @@
+'''
+Example usage of MvsUtils.py for reading MVS interface archive content.
+
+usage: MvsReadMVS.py [-h] [--input INPUT] [--output OUTPUT]
+'''
+
+from argparse import ArgumentParser
+import json
+from MvsUtils import loadMVSInterface
+import os
+
+def main():
+  parser = ArgumentParser()
+  parser.add_argument('-i', '--input', type=str, required=True, help='Path to the MVS interface archive')
+  parser.add_argument('-o', '--output', type=str, required=True, help='Path to the output json file')
+  args = parser.parse_args()
+
+  mvs = loadMVSInterface(args.input)
+
+  for platform_index in range(len(mvs['platforms'])):
+    for camera_index in range(len(mvs['platforms'][platform_index]['cameras'])):
+      camera = mvs['platforms'][platform_index]['cameras'][camera_index]
+      image_max = max(camera['width'], camera['height'])
+      fx = camera['K'][0][0] / image_max
+      fy = camera['K'][1][1] / image_max
+      poses_size = len(camera['poses'])
+      print('Camera model loaded: platform {}; camera {}; f {:.3f}x{:.3f}; poses {}'.format(platform_index, camera_index, fx, fy, poses_size))
+
+  os.makedirs(os.path.dirname(args.output), exist_ok = True)
+
+  with open(args.output, 'w') as file:
+    json.dump(mvs, file, indent=2)
+
+if __name__ == '__main__':
+  main()

scripts/python/MvsUtils.py

@@ -1,7 +1,7 @@
 '''
 OpenMVS python utilities.
 
-E.g., from MvsUtils import loadDMAP
+E.g., from MvsUtils import loadDMAP, loadMVSInterface
 '''
 
 import numpy as np
@@ -21,47 +21,170 @@ def loadDMAP(dmap_path):
     depth_width, depth_height = np.frombuffer(dmap.read(8), dtype=np.dtype('I'))
 
     if (file_type != 'DR' or has_depth == False or depth_width <= 0 or depth_height <= 0 or image_width < depth_width or image_height < depth_height):
-      print('error: opening file \'%s\' for reading depth-data' % dmap_path)
-      return None
+      print('error: opening file \'{}\' for reading depth-data'.format(dmap_path))
+      return
 
     depth_min, depth_max = np.frombuffer(dmap.read(8), dtype=np.dtype('f'))
 
-    file_name_length = np.frombuffer(dmap.read(2), dtype=np.dtype('H'))[0]
-    file_name = dmap.read(file_name_length).decode()
+    file_name_size = np.frombuffer(dmap.read(2), dtype=np.dtype('H'))[0]
+    file_name = dmap.read(file_name_size).decode()
 
-    view_ids_length = np.frombuffer(dmap.read(4), dtype=np.dtype('I'))[0]
-    reference_view_id, *neighbor_view_ids = np.frombuffer(dmap.read(4 * view_ids_length), dtype=np.dtype('I'))
+    view_ids_size = np.frombuffer(dmap.read(4), dtype=np.dtype('I'))[0]
+    reference_view_id, *neighbor_view_ids = np.frombuffer(dmap.read(4 * view_ids_size), dtype=np.dtype('I'))
 
     K = np.frombuffer(dmap.read(72), dtype=np.dtype('d')).reshape(3, 3)
     R = np.frombuffer(dmap.read(72), dtype=np.dtype('d')).reshape(3, 3)
     C = np.frombuffer(dmap.read(24), dtype=np.dtype('d'))
 
-    depth_length = depth_width * depth_height
-    depth_map = np.frombuffer(dmap.read(4 * depth_length), dtype=np.dtype('f')).reshape(depth_height, depth_width)
-    normal_map = np.frombuffer(dmap.read(4 * depth_length * 3), dtype=np.dtype('f')).reshape(depth_height, depth_width, 3) if has_normal else np.asarray([])
-    confidence_map = np.frombuffer(dmap.read(4 * depth_length), dtype=np.dtype('f')).reshape(depth_height, depth_width) if has_conf else np.asarray([])
-    views_map = np.frombuffer(dmap.read(depth_length * 4), dtype=np.dtype('B')).reshape(depth_height, depth_width, 4) if has_views else np.asarray([])
+    data = {
+      'has_normal': has_normal,
+      'has_conf': has_conf,
+      'has_views': has_views,
+      'image_width': image_width,
+      'image_height': image_height,
+      'depth_width': depth_width,
+      'depth_height': depth_height,
+      'depth_min': depth_min,
+      'depth_max': depth_max,
+      'file_name': file_name,
+      'reference_view_id': reference_view_id,
+      'neighbor_view_ids': neighbor_view_ids,
+      'K': K,
+      'R': R,
+      'C': C
+    }
+
+    map_size = depth_width * depth_height
+    depth_map = np.frombuffer(dmap.read(4 * map_size), dtype=np.dtype('f')).reshape(depth_height, depth_width)
+    data.update({'depth_map': depth_map})
+    if has_normal:
+      normal_map = np.frombuffer(dmap.read(4 * map_size * 3), dtype=np.dtype('f')).reshape(depth_height, depth_width, 3)
+      data.update({'normal_map': normal_map})
+    if has_conf:
+      confidence_map = np.frombuffer(dmap.read(4 * map_size), dtype=np.dtype('f')).reshape(depth_height, depth_width)
+      data.update({'confidence_map': confidence_map})
+    if has_views:
+      views_map = np.frombuffer(dmap.read(map_size * 4), dtype=np.dtype('B')).reshape(depth_height, depth_width, 4)
+      data.update({'views_map': views_map})
 
-  data = {
-    'has_normal': has_normal,
-    'has_conf': has_conf,
-    'has_views': has_views,
-    'image_width': image_width,
-    'image_height': image_height,
-    'depth_width': depth_width,
-    'depth_height': depth_height,
-    'depth_min': depth_min,
-    'depth_max': depth_max,
-    'file_name': file_name,
-    'reference_view_id': reference_view_id,
-    'neighbor_view_ids': neighbor_view_ids,
-    'K': K,
-    'R': R,
-    'C': C,
-    'depth_map': depth_map,
-    'normal_map': normal_map,
-    'confidence_map': confidence_map,
-    'views_map': views_map
-  }
+  return data
+
+def loadMVSInterface(archive_path):
+  with open(archive_path, 'rb') as mvs:
+    archive_type = mvs.read(4).decode()
+    version = np.frombuffer(mvs.read(4), dtype=np.dtype('I')).tolist()[0]
+    reserve = np.frombuffer(mvs.read(4), dtype=np.dtype('I'))
+
+    if archive_type != 'MVSI':
+      print('error: opening file \'{}\''.format(archive_path))
+      return
+
+    data = {
+      'project_stream': archive_type,
+      'project_stream_version': version,
+      'platforms': [],
+      'images': [],
+      'vertices': [],
+      'vertices_normal': [],
+      'vertices_color': []
+    }
+
+    platforms_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+    for platform_index in range(platforms_size):
+      platform_name_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+      platform_name = mvs.read(platform_name_size).decode()
+      data['platforms'].append({'name': platform_name, 'cameras': []})
+      cameras_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+      for camera_index in range(cameras_size):
+        camera_name_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+        camera_name = mvs.read(camera_name_size).decode()
+        data['platforms'][platform_index]['cameras'].append({'name': camera_name})
+        if version > 3:
+          band_name_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+          band_name = mvs.read(band_name_size).decode()
+          data['platforms'][platform_index]['cameras'][camera_index].update({'band_name': band_name})
+        if version > 0:
+          width, height = np.frombuffer(mvs.read(8), dtype=np.dtype('I')).tolist()
+          data['platforms'][platform_index]['cameras'][camera_index].update({'width': width, 'height': height})
+        K = np.asarray(np.frombuffer(mvs.read(72), dtype=np.dtype('d'))).reshape(3, 3).tolist()
+        data['platforms'][platform_index]['cameras'][camera_index].update({'K': K, 'poses': []})
+        identity_matrix = np.asarray(np.frombuffer(mvs.read(96), dtype=np.dtype('d'))).reshape(4, 3)
+        poses_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+        for _ in range(poses_size):
+          R = np.asarray(np.frombuffer(mvs.read(72), dtype=np.dtype('d'))).reshape(3, 3).tolist()
+          C = np.asarray(np.frombuffer(mvs.read(24), dtype=np.dtype('d'))).tolist()
+          data['platforms'][platform_index]['cameras'][camera_index]['poses'].append({'R': R, 'C': C})
+
+    images_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+    for image_index in range(images_size):
+      name_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+      name = mvs.read(name_size).decode()
+      data['images'].append({'name': name})
+      if version > 4:
+        mask_name_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+        mask_name = mvs.read(mask_name_size).decode()
+        data['images'][image_index].update({'mask_name': mask_name})
+      platform_id, camera_id, pose_id = np.frombuffer(mvs.read(12), dtype=np.dtype('I')).tolist()
+      data['images'][image_index].update({'platform_id': platform_id, 'camera_id': camera_id, 'pose_id': pose_id})
+      if version > 2:
+        id = np.frombuffer(mvs.read(4), dtype=np.dtype('I')).tolist()[0]
+        data['images'][image_index].update({'id': id})
+      if version > 6:
+        min_depth, avg_depth, max_depth = np.frombuffer(mvs.read(12), dtype=np.dtype('f')).tolist()
+        data['images'][image_index].update({'min_depth': min_depth, 'avg_depth': avg_depth, 'max_depth': max_depth, 'view_scores': []})
+        view_score_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+        for _ in range(view_score_size):
+          id, points = np.frombuffer(mvs.read(8), dtype=np.dtype('I')).tolist()
+          scale, angle, area, score = np.frombuffer(mvs.read(16), dtype=np.dtype('f')).tolist()
+          data['images'][image_index]['view_scores'].append({'id': id, 'points': points, 'scale': scale, 'angle': angle, 'area': area, 'score': score})
+
+    vertices_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+    for vertex_index in range(vertices_size):
+      X = np.frombuffer(mvs.read(12), dtype=np.dtype('f')).tolist()
+      data['vertices'].append({'X': X, 'views': []})
+      views_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+      for _ in range(views_size):
+        image_id = np.frombuffer(mvs.read(4), dtype=np.dtype('I')).tolist()[0]
+        confidence = np.frombuffer(mvs.read(4), dtype=np.dtype('f')).tolist()[0]
+        data['vertices'][vertex_index]['views'].append({'image_id': image_id, 'confidence': confidence})
+
+    vertices_normal_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+    for _ in range(vertices_normal_size):
+      normal = np.frombuffer(mvs.read(12), dtype=np.dtype('f')).tolist()
+      data['vertices_normal'].append(normal)
+
+    vertices_color_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+    for _ in range(vertices_color_size):
+      color = np.frombuffer(mvs.read(3), dtype=np.dtype('B')).tolist()
+      data['vertices_color'].append(color)
+
+    if version > 0:
+      data.update({'lines': [], 'lines_normal': [], 'lines_color': []})
+      lines_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+      for line_index in range(lines_size):
+        pt1 = np.frombuffer(mvs.read(12), dtype=np.dtype('f')).tolist()
+        pt2 = np.frombuffer(mvs.read(12), dtype=np.dtype('f')).tolist()
+        data['lines'].append({'pt1': pt1, 'pt2': pt2, 'views': []})
+        views_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+        for _ in range(views_size):
+          image_id = np.frombuffer(mvs.read(4), dtype=np.dtype('I')).tolist()[0]
+          confidence = np.frombuffer(mvs.read(4), dtype=np.dtype('f')).tolist()[0]
+          data['lines'][line_index]['views'].append({'image_id': image_id, 'confidence': confidence})
+      lines_normal_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+      for _ in range(lines_normal_size):
+        normal = np.frombuffer(mvs.read(12), dtype=np.dtype('f')).tolist()
+        data['lines_normal'].append(normal)
+      lines_color_size = np.frombuffer(mvs.read(8), dtype=np.dtype('Q'))[0]
+      for _ in range(lines_color_size):
+        color = np.frombuffer(mvs.read(3), dtype=np.dtype('B')).tolist()
+        data['lines_color'].append(color)
+      if version > 1:
+        transform = np.frombuffer(mvs.read(128), dtype=np.dtype('d')).reshape(4, 4).tolist()
+        data.update({'transform': transform})
+        if version > 5:
+          rot = np.frombuffer(mvs.read(72), dtype=np.dtype('d')).reshape(3, 3).tolist()
+          pt_min = np.frombuffer(mvs.read(24), dtype=np.dtype('d')).tolist()
+          pt_max = np.frombuffer(mvs.read(24), dtype=np.dtype('d')).tolist()
+          data.update({'obb': {'rot': rot, 'pt_min': pt_min, 'pt_max': pt_max}})
 
   return data