fast mode

Author: Jianhao-zheng

configs/wildgs_slam.yaml

@@ -8,7 +8,7 @@ device: "cuda:0"
 mapping:
   online_plotting: False # render and save images online
   full_resolution: False # if using the full resolution for mapping, but we always keep downsampled size for tracking
-  final_refine_iters: 20000 # iterations of final refinement
+  final_refine_iters: 3000 # iterations of final refinement
   eval_before_final_ba: True
   deform_gaussians: True # apply transformation on Gaussians to account for loop closure and BA
   pcd_downsample: 32 # downsamples the unprojected depth map --> point cloud

src/frontend.py

@@ -61,8 +61,8 @@ def __update(self, force_to_add_keyframe):
         for itr in range(self.iters1):
             self.graph.update(None, None, use_inactive=True)
 
-            if itr == 1 and self.video.metric_depth_reg and self.cfg['tracking']["uncertainty_params"]['activate']:
-                self.video.filter_high_err_mono_depth(self.t1-1,self.graph.ii,self.graph.jj)
+            # if itr == 1 and self.video.metric_depth_reg and self.cfg['tracking']["uncertainty_params"]['activate']:
+            #     self.video.filter_high_err_mono_depth(self.t1-1,self.graph.ii,self.graph.jj)
 
         d = self.video.distance([self.t1-2], [self.t1-1], beta=self.beta, bidirectional=True)
         # Ssee self.max_consecutive_drop_of_keyframes in initi for explanation of the following process

src/mapper.py

@@ -241,9 +241,12 @@ def run(self):
             self.keyframe_optimizers = torch.optim.Adam(opt_params)
 
             with Lock():
-                gaussian_split = self.map_opt_online(
-                    self.current_window, iters=self.mapping_itr_num
-                )
+                if video_idx % 4 == 0:
+                    gaussian_split = self.map_opt_online(
+                        self.current_window, iters=self.mapping_itr_num
+                    )
+                else:
+                    self._update_occ_aware_visibility(self.current_window)
                 if gaussian_split:
                     # do one more iteration after densify and prune
                     self.map_opt_online(self.current_window, iters=1)
@@ -546,6 +549,20 @@ def _update_mapping_points(
                 scales, "scaling"
             )["scaling"]
 
+    def _update_occ_aware_visibility(self, current_window):
+        self.occ_aware_visibility = {}
+        for kf_idx in current_window:
+            viewpoint = self.cameras[kf_idx]
+            render_pkg = render(
+                viewpoint,
+                self.gaussians,
+                self.pipeline_params,
+                self.background,
+            )
+            self.occ_aware_visibility[kf_idx] = (
+                render_pkg["n_touched"] > 0
+            ).long()
+
     def get_w2c_and_depth(self, video_idx, idx, mono_depth, print_info=False):
         est_frontend_depth, valid_depth_mask, c2w = self.video.get_depth_and_pose(
             video_idx, self.device
@@ -1146,18 +1163,7 @@ def map_opt_online(self, current_window, iters=1):
             # Deinsifying / Pruning Gaussians
             with torch.no_grad():
                 if cur_iter == iters - 1:
-                    self.occ_aware_visibility = {}
-                    for kf_idx in current_window:
-                        viewpoint = self.cameras[kf_idx]
-                        render_pkg = render(
-                            viewpoint,
-                            self.gaussians,
-                            self.pipeline_params,
-                            self.background,
-                        )
-                        self.occ_aware_visibility[kf_idx] = (
-                            render_pkg["n_touched"] > 0
-                        ).long()
+                    self._update_occ_aware_visibility(current_window)
 
                 self.gaussians.max_radii2D[visibility_filter] = torch.max(
                     self.gaussians.max_radii2D[visibility_filter],

src/utils/eval_traj.py

@@ -145,13 +145,13 @@ def full_traj_eval(traj_filler, mapper, plot_parent_dir, plot_name, stream,logge
     traj_est_lietorch = traj_est_inv.inv()
     traj_est = traj_est_lietorch.matrix().data.cpu().numpy()
 
-    ## refine non-keyframe-traj from the mapping
-    for i in tqdm(range(traj_est.shape[0])):
-        img_feat = dino_feats[i]
-        w2c_refined = mapper.refine_pose_non_key_frame(i,
-                                                    torch.tensor(np.linalg.inv(traj_est[i])),
-                                                    features=img_feat)
-        traj_est[i] = np.linalg.inv(w2c_refined.cpu().numpy())
+    # ## refine non-keyframe-traj from the mapping
+    # for i in tqdm(range(traj_est.shape[0])):
+    #     img_feat = dino_feats[i]
+    #     w2c_refined = mapper.refine_pose_non_key_frame(i,
+    #                                                 torch.tensor(np.linalg.inv(traj_est[i])),
+    #                                                 features=img_feat)
+    #     traj_est[i] = np.linalg.inv(w2c_refined.cpu().numpy())
 
     kf_num = traj_filler.video.counter.value
     kf_timestamps = traj_filler.video.timestamp[:kf_num].cpu().int().numpy()