Adds option for RGB rendering

README.md

@@ -124,7 +124,7 @@ The latest model is available to download: [[checkpoint](https://drive.google.co
 This code belongs to Robotic Systems Lab, ETH Zurich.
 All right reserved
 
-**Authors: [Pascal Roth](https://github.com/pascal-roth), [Julian Nubert](https://juliannubert.com/), [Fan Yang](https://github.com/MichaelFYang), [Mayank Mittal](https://mayankm96.github.io/), [Ziqi Fan](https://github.com/fan-ziqi), and [Marco Hutter](https://rsl.ethz.ch/the-lab/people/person-detail.MTIxOTEx.TGlzdC8yNDQxLC0xNDI1MTk1NzM1.html)<br />
+**Authors: [Pascal Roth](https://github.com/pascal-roth), [Julian Nubert](https://juliannubert.com/), [Fan Yang](https://github.com/MichaelFYang), [Mayank Mittal](https://mayankm96.github.io/), and [Marco Hutter](https://rsl.ethz.ch/the-lab/people/person-detail.MTIxOTEx.TGlzdC8yNDQxLC0xNDI1MTk1NzM1.html)<br />
 Maintainer: Pascal Roth, [email protected]**
 
 The ViPlanner package has been tested under ROS Noetic on Ubuntu 20.04.

omniverse/extension/omni.viplanner/omni/viplanner/collectors/viewpoint_sampling.py

@@ -81,9 +81,9 @@ def sample_viewpoints(self, nbr_viewpoints: int, seed: int = 1) -> torch.Tensor:
             sample_idx_select = torch.randperm(sample_idx.sum())[
                 : min(nbr_samples_per_point, nbr_viewpoints - sample_locations_count)
             ]
-            sample_locations[
-                sample_locations_count : sample_locations_count + sample_idx_select.shape[0]
-            ] = self.terrain_analyser.samples[sample_idx][sample_idx_select, :2]
+            sample_locations[sample_locations_count : sample_locations_count + sample_idx_select.shape[0]] = (
+                self.terrain_analyser.samples[sample_idx][sample_idx_select, :2]
+            )
             sample_locations_count += sample_idx_select.shape[0]
             curr_point_idx += 1
             # reset point index if all points are sampled
@@ -157,6 +157,8 @@ def render_viewpoints(self, samples: torch.Tensor):
         # create directories
         os.makedirs(os.path.join(filedir, "semantics"), exist_ok=True)
         os.makedirs(os.path.join(filedir, "depth"), exist_ok=True)
+        if "rgb" in self.cfg.cameras.values():
+            os.makedirs(os.path.join(filedir, "rgb"), exist_ok=True)
 
         # save camera configurations
         print(f"[INFO] Saving camera configurations to {filedir}.")
@@ -206,16 +208,18 @@ def render_viewpoints(self, samples: torch.Tensor):
                 # save images
                 for idx in range(samples_idx.shape[0]):
                     # semantic segmentation
-                    if annotator == "semantic_segmentation":
+                    if annotator == "semantic_segmentation" or annotator == "rgb":
                         if image_data_np.shape[-1] == 1:
                             # get info data
                             info = self.scene.sensors[cam].data.info[idx][annotator]["idToLabels"]
 
                             # assign each key a color from the VIPlanner color space
                             info = {
-                                int(k): self.viplanner_sem_meta.class_color["static"]
-                                if v["class"] in ("BACKGROUND", "UNLABELLED")
-                                else self.viplanner_sem_meta.class_color[v["class"]]
+                                int(k): (
+                                    self.viplanner_sem_meta.class_color["static"]
+                                    if v["class"] in ("BACKGROUND", "UNLABELLED")
+                                    else self.viplanner_sem_meta.class_color[v["class"]]
+                                )
                                 for k, v in info.items()
                             }
 
@@ -232,7 +236,11 @@ def render_viewpoints(self, samples: torch.Tensor):
                             output = image_data_np[idx]
 
                         assert cv2.imwrite(
-                            os.path.join(filedir, "semantics", f"{image_idx[cam_idx]}".zfill(4) + ".png"),
+                            os.path.join(
+                                filedir,
+                                "semantics" if annotator == "semantic_segmentation" else "rgb",
+                                f"{image_idx[cam_idx]}".zfill(4) + ".png"
+                            ),
                             cv2.cvtColor(output.astype(np.uint8), cv2.COLOR_RGB2BGR),
                         )
                     # depth

omniverse/extension/omni.viplanner/omni/viplanner/config/carla_cfg.py

@@ -82,6 +82,8 @@ class TerrainSceneCfg(InteractiveSceneCfg):
         height=480,
         data_types=["distance_to_image_plane"],
     )
+
+    # NOTE: remove "rgb" from the data_types to only render the semantic segmentation
     semantic_camera = CameraCfg(
         prim_path="{ENV_REGEX_NS}/Robot/base/semantic_camera",
         offset=CameraCfg.OffsetCfg(pos=(0.510, 0.0, 0.015), rot=(-0.5, 0.5, -0.5, 0.5)),
@@ -91,7 +93,7 @@ class TerrainSceneCfg(InteractiveSceneCfg):
         ),
         width=1280,
         height=720,
-        data_types=["semantic_segmentation"],
+        data_types=["semantic_segmentation", "rgb"],
         colorize_semantic_segmentation=False,
     )
 

omniverse/extension/omni.viplanner/omni/viplanner/config/matterport_cfg.py

@@ -11,7 +11,7 @@
 from omni.isaac.lab.managers import ObservationTermCfg as ObsTerm
 from omni.isaac.lab.managers import SceneEntityCfg
 from omni.isaac.lab.scene import InteractiveSceneCfg
-from omni.isaac.lab.sensors import ContactSensorCfg, patterns
+from omni.isaac.lab.sensors import ContactSensorCfg, patterns, CameraCfg
 from omni.isaac.lab.utils import configclass
 from omni.isaac.matterport.config import MatterportImporterCfg
 from omni.isaac.matterport.domains import MatterportRayCasterCfg
@@ -121,6 +121,17 @@ class TerrainSceneCfg(InteractiveSceneCfg):
         debug_vis=False,
         mesh_prim_paths=["${USER_PATH_TO_USD}/matterport.ply"],
     )
+
+    # NOTE: comment the following lines to only render the semantic segmentation and depth images
+    rgb_camera = CameraCfg(
+        prim_path="{ENV_REGEX_NS}/Robot/base/rgb_camera",
+        offset=CameraCfg.OffsetCfg(pos=(0.510, 0.0, 0.015), rot=(-0.5, 0.5, -0.5, 0.5)),
+        spawn=sim_utils.PinholeCameraCfg(),
+        width=1280,
+        height=720,
+        data_types=["rgb"],
+        colorize_semantic_segmentation=False,
+    )
 
 
 @configclass

omniverse/extension/omni.viplanner/omni/viplanner/config/warehouse_cfg.py

@@ -80,13 +80,15 @@ class TerrainSceneCfg(InteractiveSceneCfg):
         height=480,
         data_types=["distance_to_image_plane"],
     )
+
+    # NOTE: remove "rgb" from the data_types to only render the semantic segmentation
     semantic_camera = CameraCfg(
         prim_path="{ENV_REGEX_NS}/Robot/base/semantic_camera",
         offset=CameraCfg.OffsetCfg(pos=(0.510, 0.0, 0.015), rot=(-0.5, 0.5, -0.5, 0.5)),
         spawn=sim_utils.PinholeCameraCfg(),
         width=1280,
         height=720,
-        data_types=["semantic_segmentation"],
+        data_types=["semantic_segmentation", "rgb"],
         colorize_semantic_segmentation=False,
     )
 

omniverse/standalone/data_collect.py

@@ -97,6 +97,13 @@ def main():
         "semantic_camera": "semantic_segmentation",
     }
 
+    # adustments if also RGB images should be rendered
+    if args_cli.scene == "matterport" and hasattr(scene_cfg, "rgb_camera"):
+        scene_cfg.rgb_camera.prim_path = "/World/rgb_camera"
+        cfg.cameras["rgb_camera"] = "rgb"
+    elif "rgb" in scene_cfg.semantic_camera.data_types:
+        cfg.cameras["semantic_camera"] = "rgb"
+
     # Load kit helper
     sim_cfg = sim_utils.SimulationCfg()
     sim = SimulationContext(sim_cfg)