handle indoor case

* To do : use flight data to deduce model scale

cli_utils.py

@@ -54,6 +54,7 @@ def set_argparser():
     exec_parser.add_argument("--pcl_util", default="pcl_util/build", type=Path)
 
     pm_parser = parser.add_argument_group("Photogrammetry")
+    pm_parser.add_argument('--max_num_matches', default=32768, type=int, help="max number of matches, lower it if you get GPU memory error")
     pm_parser.add_argument('--vocab_tree', type=Path, default="vocab_tree_flickr100K_words256K.bin")
     pm_parser.add_argument('--triangulate', action="store_true")
     pm_parser.add_argument('--multiple_models', action='store_true', help='If selected, will let colmap mapper do multiple models.'

main_pipeline.py

@@ -73,6 +73,8 @@ def main():
     no_gt_folder = args.input_folder/"Videos"/"no_groundtruth"
     if no_gt_folder.isdir():
         env["videos_to_localize"] = [v for v in env["videos_list"] if not str(v).startswith(no_gt_folder)]
+    else:
+        env["videos_to_localize"] = env["videos_list"]
 
     i = 1
     if i not in args.skip_step:
@@ -114,7 +116,7 @@ def main():
         env["thorough_recon"].makedirs_p()
         colmap.extract_features(image_list=env["video_frame_list_thorough"], more=args.more_sift_features)
         colmap.index_images(vocab_tree_output=env["indexed_vocab_tree"], vocab_tree_input=args.vocab_tree)
-        colmap.match(method="vocab_tree", vocab_tree=env["indexed_vocab_tree"])
+        colmap.match(method="vocab_tree", vocab_tree=env["indexed_vocab_tree"], max_num_matches=env["max_num_matches"])
         colmap.map(output=env["thorough_recon"], multiple_models=env["multiple_models"])
         thorough_model = pi.choose_biggest_model(env["thorough_recon"])
         colmap.adjust_bundle(thorough_model, thorough_model,
@@ -132,7 +134,7 @@ def main():
         if not (env["georef_frames_list"]/"images.bin").isfile():
             # GPS alignment failed, possibly because not enough GPS referenced images
             # Copy the original model without alignment
-            (env["thorough_recon"] / "0").merge_tree(env["georef_full_recon"])
+            (env["thorough_recon"] / "0").merge_tree(env["georef_recon"])
         env["georef_recon"].merge_tree(env["georef_full_recon"])
     if args.inspect_dataset:
         colmap.export_model(output=env["georef_recon"] / "georef_sparse.ply",

main_pipeline_no_lidar.py

@@ -72,7 +72,7 @@ def main():
         env["thorough_recon"].makedirs_p()
         colmap.extract_features(image_list=env["video_frame_list_thorough"], more=args.more_sift_features)
         colmap.index_images(vocab_tree_output=env["indexed_vocab_tree"], vocab_tree_input=args.vocab_tree)
-        colmap.match(method="vocab_tree", vocab_tree=env["indexed_vocab_tree"])
+        colmap.match(method="vocab_tree", vocab_tree=env["indexed_vocab_tree"], max_num_matches=env["max_num_matches"])
         colmap.map(output=env["thorough_recon"], multiple_models=env["multiple_models"])
         thorough_model = pi.choose_biggest_model(env["thorough_recon"])
         colmap.adjust_bundle(thorough_model, thorough_model,
@@ -90,7 +90,7 @@ def main():
         if not (env["georef_frames_list"]/"images.bin").isfile():
             # GPS alignment failed, possibly because not enough GPS referenced images
             # Copy the original model without alignment
-            (env["thorough_recon"] / "0").merge_tree(env["georef_full_recon"])
+            (env["thorough_recon"] / "0").merge_tree(env["georef_recon"])
         env["georef_recon"].merge_tree(env["georef_full_recon"])
     if args.inspect_dataset:
         colmap.export_model(output=env["georef_recon"] / "georef_sparse.ply",

video_localization.py

@@ -19,12 +19,12 @@ def is_video_in_model(video_name, colmap_model, metadata):
     return sum(video_image_ids.isin(mapped_images_ids)) > 0
 
 
-def sift_and_match(colmap, more_sift_features, indexed_vocab_tree, image_list, **env):
+def sift_and_match(colmap, more_sift_features, indexed_vocab_tree, image_list, max_num_matches, **env):
     tries = 0
     while tries < 10:
         try:
             colmap.extract_features(image_list=image_list, more=more_sift_features)
-            colmap.match(method="sequential", vocab_tree=indexed_vocab_tree)
+            colmap.match(method="sequential", vocab_tree=indexed_vocab_tree, max_num_matches=max_num_matches)
         except Exception:
             # If it failed, that's because sift gpu has failed.
             print("Error With colmap, will retry")

wrappers/colmap.py

@@ -29,10 +29,11 @@ class Colmap(Wrapper):
             pass
         self.__call__(options)
 
-    def match(self, method="exhaustive", guided_matching=True, vocab_tree=None):
+    def match(self, method="exhaustive", guided_matching=True, vocab_tree=None, max_num_matches=32768):
         options = ["{}_matcher".format(method),
                    "--database_path", self.db,
-                   "--SiftMatching.guided_matching", "1" if guided_matching else "0"]
+                   "--SiftMatching.guided_matching", "1" if guided_matching else "0",
+                   "--SiftMatching.max_num_matches", str(max_num_matches)]
         if method == "sequential":
             assert vocab_tree is not None
             options += ["--SequentialMatching.loop_detection", "1",