Fix typos

datasets/phototourism.py

@@ -18,7 +18,7 @@
 from kornia import create_meshgrid
 import h5py
 
-# additinal configuarion
+# additional configuration
 sfm_path = "sparse"
 vis_octree = False
 vis_intersection = False
@@ -274,7 +274,7 @@ def near_far_voxel(self, octree_data, rays_o, rays_d, image_name, chunk_size=655
         voxel_near_sfm_all = []
         voxel_far_sfm_all = []
 
-        # todo: figure out why chunck size greater or equal than 1768500 will result in error intersection
+        # todo: figure out why chunk size greater or equal than 1768500 will result in error intersection
         # use 1000000 as a threshold just to be safe
         chunk_size = min(rays_o.size()[0], 100000)
         try:
@@ -663,9 +663,9 @@ def read_meta(self):
 
                         valid_num = torch.sum(valid_depth).long().item()
                         current_len = rays.size()[0]
-                        curent_percent = valid_num / current_len
+                        current_percent = valid_num / current_len
                         padding_length = int(np.ceil((self.depth_percent * current_len - valid_num) / (1 - self.depth_percent)))
-                        print(f"padding valid depth percentage: from {curent_percent} to {self.depth_percent} with padding {padding_length}")
+                        print(f"padding valid depth percentage: from {current_percent} to {self.depth_percent} with padding {padding_length}")
 
                         pad_ind =  torch.floor((torch.rand(padding_length) * valid_num)).long()
                         result_length = padding_length + current_len

lightning_modules/neuconw_system.py

@@ -106,7 +106,7 @@ def __init__(self, hparams, config, caches):
 
         spc_options = {
             "voxel_size": self.scene_config["voxel_size"],
-            "recontruct_path": self.config.DATASET.ROOT_DIR,
+            "reconstruct_path": self.config.DATASET.ROOT_DIR,
             "min_track_length": self.scene_config["min_track_length"],
         }
         self.renderer = NeuconWRenderer(
@@ -272,14 +272,14 @@ def octree_update(
 
         del self.renderer.fine_octree_data
 
-        # get suface points
+        # get surface points
         sparse_pc_sfm, train_voxel_size = self.surface_selection(
             train_level, threshold, device, chunk
         )
 
         # use remaining points to generate new octree
         octree_new, scene_origin, scale, level = gen_octree(
-            self.renderer.recontruct_path,
+            self.renderer.reconstruct_path,
             sparse_pc_sfm,
             train_voxel_size,
             device=device,

rendering/renderer.py

@@ -94,14 +94,14 @@ def __init__(
         self.sample_range = sample_range
         self.fine_octree_data = None
         if self.nerf_far_override:
-            self.recontruct_path = spc_options["recontruct_path"]
+            self.reconstruct_path = spc_options["reconstruct_path"]
             self.min_track_length = spc_options["min_track_length"]
             self.voxel_size = spc_options["voxel_size"]
 
         self.sfm_to_gt = np.eye(4)
 
         # read unit sphere origin and radius from scene config
-        scene_config_path = os.path.join(spc_options["recontruct_path"], "config.yaml")
+        scene_config_path = os.path.join(spc_options["reconstruct_path"], "config.yaml")
         if os.path.isfile(scene_config_path):
             with open(scene_config_path, "r") as yamlfile:
                 scene_config = yaml.load(yamlfile, Loader=yaml.FullLoader)
@@ -136,7 +136,7 @@ def __init__(
 
     def get_octree(self, device):
         octree, scene_origin, scale, level = gen_octree_from_sfm(
-            self.recontruct_path, self.min_track_length, self.voxel_size, device=device
+            self.reconstruct_path, self.min_track_length, self.voxel_size, device=device
         )
 
         octree_data = {}
@@ -385,7 +385,7 @@ def get_near_far_octree(self, octree_data, rays_o, rays_d, near, far):
         octree_level = octree_data["level"]
         spc_data = octree_data["spc_data"]
 
-        # transfrom origins and direction of rays to sfm coordinate system
+        # transform origins and direction of rays to sfm coordinate system
         rays_o_sfm = (rays_o * self.radius).view(-1, 3) + self.origin
 
         # generate near far from spc
@@ -423,7 +423,7 @@ def get_near_far_sdf(self, octree_data, rays_o, rays_d, near, far):
         train_voxel_size = octree_data["voxel_size"]
         train_spc_data = octree_data["spc_data"]
 
-        # transfrom origins and direction of rays to sfm coordinate system
+        # transform origins and direction of rays to sfm coordinate system
         rays_o_sfm = (rays_o * self.radius).view(-1, 3) + self.origin
 
         # generate near far from spc
@@ -456,7 +456,7 @@ def get_near_far_sdf(self, octree_data, rays_o, rays_d, near, far):
         return voxel_near, voxel_far, ~miss_mask
 
     def sparse_sampler(self, rays_o, rays_d, near, far, perturb):
-        """sample on spaese voxel. Including upsample on sparse voxels,
+        """sample on sparse voxel. Including upsample on sparse voxels,
         and uniform sample on inverse depth of original near far,
         Note that input coordinates are scaled to unit sphere
 

tools/extract_mesh.py

@@ -29,13 +29,13 @@ def get_opts():
                         choices=['blender', 'phototourism'],
                         help='which dataset to validate')
     parser.add_argument('--eval_level', type=int, default=-1,
-                        help='level og precision')
+                        help='level of precision')
     parser.add_argument('--mesh_size', type=int, default=128,
                         help='resolution of mesh, (N, N, N)')
     parser.add_argument('--mesh_origin', type=str, default="0, 0, 0",
                         help='origin of mesh, (x, y, z)')
     parser.add_argument('--mesh_radius', type=float, default=1.0,
-                        help='radius pf mesh')
+                        help='radius of mesh')
     parser.add_argument('--vertex_color', default=False, action="store_true",
                         help='whether add color to mesh')
     parser.add_argument('--num_gpus', type=int, default=1,

tools/prepare_data/generate_voxel.py

@@ -39,7 +39,7 @@ def expand_points(points, voxel_size):
 
 
 def gen_octree_from_sfm(
-    recontruct_path,
+    reconstruct_path,
     min_track_length,
     voxel_size,
     sfm_path="sparse",
@@ -49,7 +49,7 @@ def gen_octree_from_sfm(
     radius=1.0,
 ):
     # read 3d points from sfm result, and filter them
-    point_path = os.path.join(recontruct_path, f"dense/{sfm_path}/points3D.bin")
+    point_path = os.path.join(reconstruct_path, f"dense/{sfm_path}/points3D.bin")
     points_3d = read_points3d_binary(point_path)
     points_ori = []
     for id, p in points_3d.items():
@@ -68,12 +68,12 @@ def gen_octree_from_sfm(
         o3d.io.write_point_cloud(f"samples/voxel_vis_source.ply", gt_pcd)
 
     return gen_octree(
-        recontruct_path, points, voxel_size, device, visualize, expand, radius
+        reconstruct_path, points, voxel_size, device, visualize, expand, radius
     )
 
 
 def gen_octree(
-    recontruct_path,
+    reconstruct_path,
     points,
     voxel_size,
     device=0,
@@ -82,7 +82,7 @@ def gen_octree(
     radius=1.0,
     in_sfm=True,
 ):
-    scene_config_path = os.path.join(recontruct_path, "config.yaml")
+    scene_config_path = os.path.join(reconstruct_path, "config.yaml")
     # read scene config
     with open(scene_config_path, "r") as yamlfile:
         scene_config = yaml.load(yamlfile, Loader=yaml.FullLoader)
@@ -107,7 +107,7 @@ def gen_octree(
     # dimensions
     dim = np.max(bbx_max - bbx_min)
 
-    # points dialation
+    # points dilation
     for _ in range(expand):
         points = expand_points(points, voxel_size)
 
@@ -192,7 +192,7 @@ def level_upgrade(
     octree_scale,
     src_level,
     target_level,
-    recontruct_path,
+    reconstruct_path,
     visualize=False,
 ):
     # upsample octree
@@ -231,7 +231,7 @@ def level_upgrade(
     xyz_sfm = sparse_ind_up * target_voxel_size + vol_origin
 
     return gen_octree(
-        recontruct_path,
+        reconstruct_path,
         xyz_sfm.cpu().numpy(),
         target_voxel_size,
         device=device,
@@ -246,7 +246,7 @@ def level_downgrade(
     octree_scale,
     src_level,
     target_level,
-    recontruct_path,
+    reconstruct_path,
     visualize=False,
 ):
     device = octree.device
@@ -266,7 +266,7 @@ def level_downgrade(
     target_voxel_size = 2 / (2**target_level) * octree_scale
 
     return gen_octree(
-        recontruct_path,
+        reconstruct_path,
         xyz_sfm,
         target_voxel_size,
         device=device,
@@ -281,7 +281,7 @@ def octree_level_adjust(
     octree_scale,
     src_level,
     target_level,
-    recontruct_path,
+    reconstruct_path,
     visualize,
 ):
     if target_level > src_level:
@@ -291,7 +291,7 @@ def octree_level_adjust(
             octree_scale,
             src_level,
             target_level,
-            recontruct_path,
+            reconstruct_path,
             visualize,
         )
     elif target_level < src_level:
@@ -301,7 +301,7 @@ def octree_level_adjust(
             octree_scale,
             src_level,
             target_level,
-            recontruct_path,
+            reconstruct_path,
             visualize,
         )
     else:
@@ -328,7 +328,7 @@ def get_near_far(
 
     'with_exit': set true to obtain accurate far. Default to false as this will perform aabb twice
     """
-    # Avoid corner cases. issuse in kaolin: https://github.com/NVIDIAGameWorks/kaolin/issues/490
+    # Avoid corner cases. issues in kaolin: https://github.com/NVIDIAGameWorks/kaolin/issues/490
     rays_d = rays_d.clone() + 1e-7
     rays_o = rays_o.clone() + 1e-7
 
@@ -446,10 +446,10 @@ def get_near_far(
     from torch.utils.data import DataLoader
 
     voxel_size = 0.1
-    recontruct_path = "/nas/datasets/IMC/phototourism/training_set/brandenburg_gate"
+    reconstruct_path = "/nas/datasets/IMC/phototourism/training_set/brandenburg_gate"
     min_track_length = 50
     octree, scene_origin, scale, level = gen_octree_from_sfm(
-        recontruct_path, min_track_length, voxel_size, visualize=True
+        reconstruct_path, min_track_length, voxel_size, visualize=True
     )
     # gen fake ray origin and direction
     rays_o = (

tools/reproj_error.py

@@ -140,13 +140,13 @@ def image_reproj_error(imdata, pts3d, img_ids, entrinsics_dict, intrinsics_dict)
 
 
 
-def gt_reproject_error(data_dir, gt_pcd_path, sfm_to_gt, reconstuct_path, track_length=200, reproj_error=0.4, batch_size=2, img_reproj_error=300):
+def gt_reproject_error(data_dir, gt_pcd_path, sfm_to_gt, reconstruct_path, track_length=200, reproj_error=0.4, batch_size=2, img_reproj_error=300):
     # 0. read data from bin
-    # reconstuct_path = 'dense_ws_filtered_tkl200_mrep.5'
-    # reconstuct_path = 'dense/sparse'
-    imdata = read_images_binary(os.path.join(data_dir, reconstuct_path, 'images.bin'))
-    camdata = read_cameras_binary(os.path.join(data_dir, reconstuct_path, 'cameras.bin'))
-    pts3d = read_points3d_binary(os.path.join(data_dir, reconstuct_path, 'points3D.bin'))
+    # reconstruct_path = 'dense_ws_filtered_tkl200_mrep.5'
+    # reconstruct_path = 'dense/sparse'
+    imdata = read_images_binary(os.path.join(data_dir, reconstruct_path, 'images.bin'))
+    camdata = read_cameras_binary(os.path.join(data_dir, reconstruct_path, 'cameras.bin'))
+    pts3d = read_points3d_binary(os.path.join(data_dir, reconstruct_path, 'points3D.bin'))
 
     gt_pcd = torch.from_numpy(np.array(o3d.io.read_point_cloud(gt_pcd_path).points)).float().cuda()
     img_ids, img_id_to_name = get_image_id(imdata, data_dir)
@@ -254,7 +254,7 @@ def get_opts():
     parser.add_argument('--gt_pcd_path', type=str,
                         default="/nas/datasets/OpenHeritage3D/pro/brandenburg_gate/bg_sampled_0.01_cropped.ply",
                         help='target point cloud')
-    parser.add_argument('--reconstuct_path', type=str,
+    parser.add_argument('--reconstruct_path', type=str,
                         default="dense/sparse",
                         help='reconstruction work space')
     parser.add_argument('--track_length', type=int, default='200',
@@ -274,4 +274,4 @@ def get_opts():
     # read scene config
     with open(os.path.join(args.data_dir, 'config.yaml'), "r") as yamlfile:
         scene_config = yaml.load(yamlfile, Loader=yaml.FullLoader)
-    gt_reproject_error(args.data_dir, args.gt_pcd_path, np.array(scene_config['sfm2gt']), args.reconstuct_path, args.track_length, args.reproj_error, args.batch_size, args.img_reproj_error)
+    gt_reproject_error(args.data_dir, args.gt_pcd_path, np.array(scene_config['sfm2gt']), args.reconstruct_path, args.track_length, args.reproj_error, args.batch_size, args.img_reproj_error)

utils/kaolin_renderer.py

@@ -116,7 +116,7 @@ def __call__(self, height, width, intrinsic, pose):
 
         depth_all = []
         pid_all = []
-        # todo: figure out why chunck size greater or equal than 1768500 will result in error intersection
+        # todo: figure out why chunk size greater or equal than 1768500 will result in error intersection
         # use 1000000 as a threshold just to be safe
         chunk_size = min(rays_o.size()[0], 1000000)
         for i in range(0, rays_o.size()[0], chunk_size):

utils/reproj_filter.py

@@ -252,12 +252,12 @@ def split_list(_list, n):
 
 
 if __name__ == "__main__":
-    reconstuct_path = 'dense/sparse'
+    reconstruct_path = 'dense/sparse'
     print(f"result will be saved to {args.output_path}")
     os.makedirs(args.output_path, exist_ok=True)
 
-    imdata = read_images_binary(os.path.join(args.data_path, reconstuct_path, 'images.bin'))
-    camdata = read_cameras_binary(os.path.join(args.data_path, reconstuct_path, 'cameras.bin'))
+    imdata = read_images_binary(os.path.join(args.data_path, reconstruct_path, 'images.bin'))
+    camdata = read_cameras_binary(os.path.join(args.data_path, reconstruct_path, 'cameras.bin'))
     img_ids_all, img_id_to_name, img_path_to_id = get_image_id(imdata, args.data_path)
     img_ids = get_train_ids(args.data_path, img_ids_all, img_path_to_id)
     print(f"views to process: {len(img_ids)}")

utils/vis_metrics.py

@@ -18,7 +18,7 @@ def get_opts():
     return parser.parse_args()   
 
 
-def svae_plot(ind, data1, data2, name1, name2, save_path, name):
+def save_plot(ind, data1, data2, name1, name2, save_path, name):
     plt.plot(ind, np.array(data1) * 100, "-b", label=name1)
     plt.plot(ind, np.array(data2) * 100, "-r", label=name2)
     plt.legend(loc="upper left")
@@ -46,7 +46,7 @@ def vis_results(ours_path, colmap_path, save_name, max_num):
     save_path = os.path.join("eval_results", f"{save_name}")
     os.makedirs(save_path, exist_ok=True)
     for key in ours_metrics.keys():
-        svae_plot(thresholds, ours_metrics[key][:max_num],  colmap_metrics[key][:max_num], \
+        save_plot(thresholds, ours_metrics[key][:max_num],  colmap_metrics[key][:max_num], \
                     "ours", "colmap", save_path, f"{key}")
 
 if __name__ == "__main__":