Specifying views from config

code/datasets/scene_dataset.py

@@ -13,7 +13,8 @@ def __init__(self,
                  data_dir,
                  img_res,
                  scene_id='scan0',
-                 cam_file=None
+                 cam_file=None,
+                 views=None
                  ):
 
         self.instance_dir = os.path.join('../data', data_dir, '{0}'.format(scene_id))
@@ -31,15 +32,20 @@ def __init__(self,
         mask_dir = '{0}/mask'.format(self.instance_dir)
         mask_paths = sorted(utils.glob_imgs(mask_dir))
 
-        self.n_images = len(image_paths)
+        if views:
+            self.views = views
+            image_paths = [image_paths[v] for v in self.views]
+            mask_paths = [mask_paths[v] for v in self.views]
+        else:
+            self.views = list(range(len(image_paths)))
 
         self.cam_file = '{0}/cameras.npz'.format(self.instance_dir)
         if cam_file is not None:
             self.cam_file = '{0}/{1}'.format(self.instance_dir, cam_file)
 
         camera_dict = np.load(self.cam_file)
-        scale_mats = [camera_dict['scale_mat_%d' % idx].astype(np.float32) for idx in range(self.n_images)]
-        world_mats = [camera_dict['world_mat_%d' % idx].astype(np.float32) for idx in range(self.n_images)]
+        scale_mats = [camera_dict['scale_mat_%d' % idx].astype(np.float32) for idx in self.views]
+        world_mats = [camera_dict['world_mat_%d' % idx].astype(np.float32) for idx in self.views]
 
         self.intrinsics_all = []
         self.pose_all = []
@@ -63,7 +69,7 @@ def __init__(self,
             self.object_masks.append(torch.from_numpy(object_mask).bool())
 
     def __len__(self):
-        return self.n_images
+        return len(self.views)
 
     def __getitem__(self, idx):
         uv = np.mgrid[0:self.img_res[0], 0:self.img_res[1]].astype(np.int32)
@@ -119,8 +125,8 @@ def get_scale_mat(self):
     def get_gt_pose(self, scaled=False):
         # Load gt pose without normalization to unit sphere
         camera_dict = np.load(self.cam_file)
-        world_mats = [camera_dict['world_mat_%d' % idx].astype(np.float32) for idx in range(self.n_images)]
-        scale_mats = [camera_dict['scale_mat_%d' % idx].astype(np.float32) for idx in range(self.n_images)]
+        world_mats = [camera_dict['world_mat_%d' % idx].astype(np.float32) for idx in self.views]
+        scale_mats = [camera_dict['scale_mat_%d' % idx].astype(np.float32) for idx in self.views]
 
         pose_all = []
         for scale_mat, world_mat in zip(scale_mats, world_mats):
@@ -137,8 +143,8 @@ def get_pose_init(self):
         # get noisy initializations obtained with the linear method
         cam_file = '{0}/cameras_linear_init.npz'.format(self.instance_dir)
         camera_dict = np.load(cam_file)
-        scale_mats = [camera_dict['scale_mat_%d' % idx].astype(np.float32) for idx in range(self.n_images)]
-        world_mats = [camera_dict['world_mat_%d' % idx].astype(np.float32) for idx in range(self.n_images)]
+        scale_mats = [camera_dict['scale_mat_%d' % idx].astype(np.float32) for idx in self.views]
+        world_mats = [camera_dict['world_mat_%d' % idx].astype(np.float32) for idx in self.views]
 
         init_pose = []
         for scale_mat, world_mat in zip(scale_mats, world_mats):