Revert "feat(visualization): allow remote viewer + compress rerun images (#2756)" (#2766)

This reverts commit f844c7a458.

src/lerobot/scripts/lerobot_dataset_viz.py

@@ -96,7 +96,6 @@ def visualize_dataset(
     ws_port: int = 9087,
     save: bool = False,
     output_dir: Path | None = None,
-    display_compressed_images: bool = False,
 ) -> Path | None:
     if save:
         assert output_dir is not None, (
@@ -138,9 +137,8 @@ def visualize_dataset(
 
             # display each camera image
             for key in dataset.meta.camera_keys:
-                img = to_hwc_uint8_numpy(batch[key][i])
+                # TODO(rcadene): add `.compress()`? is it lossless?
-                img_entity = rr.Image(img).compress() if display_compressed_images else rr.Image(img)
+                rr.log(key, rr.Image(to_hwc_uint8_numpy(batch[key][i])))
-                rr.log(key, entity=img_entity)
 
             # display each dimension of action space (e.g. actuators command)
             if ACTION in batch:
@@ -263,14 +261,6 @@ def main():
         ),
     )
 
-    parser.add_argument(
-        "--display-compressed-images",
-        type=bool,
-        required=True,
-        default=False,
-        help="If set, display compressed images in Rerun instead of uncompressed ones.",
-    )
-
     args = parser.parse_args()
     kwargs = vars(args)
     repo_id = kwargs.pop("repo_id")

src/lerobot/scripts/lerobot_record.py

@@ -185,12 +185,6 @@ class RecordConfig:
     policy: PreTrainedConfig | None = None
     # Display all cameras on screen
     display_data: bool = False
-    # Display data on a remote Rerun server
-    display_ip: str | None = None
-    # Port of the remote Rerun server
-    display_port: int | None = None
-    # Whether to  display compressed images in Rerun
-    display_compressed_images: bool = False
     # Use vocal synthesis to read events.
     play_sounds: bool = True
     # Resume recording on an existing dataset.
@@ -267,7 +261,6 @@ def record_loop(
     control_time_s: int | None = None,
     single_task: str | None = None,
     display_data: bool = False,
-    display_compressed_images: bool = False,
 ):
     if dataset is not None and dataset.fps != fps:
         raise ValueError(f"The dataset fps should be equal to requested fps ({dataset.fps} != {fps}).")
@@ -378,9 +371,7 @@ def record_loop(
             dataset.add_frame(frame)
 
         if display_data:
-            log_rerun_data(
+            log_rerun_data(observation=obs_processed, action=action_values)
-                observation=obs_processed, action=action_values, compress_images=display_compressed_images
-            )
 
         dt_s = time.perf_counter() - start_loop_t
         precise_sleep(max(1 / fps - dt_s, 0.0))
@@ -393,12 +384,7 @@ def record(cfg: RecordConfig) -> LeRobotDataset:
     init_logging()
     logging.info(pformat(asdict(cfg)))
     if cfg.display_data:
-        init_rerun(session_name="recording", ip=cfg.display_ip, port=cfg.display_port)
+        init_rerun(session_name="recording")
-    display_compressed_images = (
-        True
-        if (cfg.display_data and cfg.display_ip is not None and cfg.display_port is not None)
-        else cfg.display_compressed_images
-    )
 
     robot = make_robot_from_config(cfg.robot)
     teleop = make_teleoperator_from_config(cfg.teleop) if cfg.teleop is not None else None
@@ -492,7 +478,6 @@ def record(cfg: RecordConfig) -> LeRobotDataset:
                     control_time_s=cfg.dataset.episode_time_s,
                     single_task=cfg.dataset.single_task,
                     display_data=cfg.display_data,
-                    display_compressed_images=display_compressed_images,
                 )
 
                 # Execute a few seconds without recording to give time to manually reset the environment

src/lerobot/scripts/lerobot_teleoperate.py

@@ -108,12 +108,6 @@ class TeleoperateConfig:
     teleop_time_s: float | None = None
     # Display all cameras on screen
     display_data: bool = False
-    # Display data on a remote Rerun server
-    display_ip: str | None = None
-    # Port of the remote Rerun server
-    display_port: int | None = None
-    # Whether to  display compressed images in Rerun
-    display_compressed_images: bool = False
 
 
 def teleop_loop(
@@ -125,7 +119,6 @@ def teleop_loop(
     robot_observation_processor: RobotProcessorPipeline[RobotObservation, RobotObservation],
     display_data: bool = False,
     duration: float | None = None,
-    display_compressed_images: bool = False,
 ):
     """
     This function continuously reads actions from a teleoperation device, processes them through optional
@@ -137,7 +130,6 @@ def teleop_loop(
         robot: The robot instance being controlled.
         fps: The target frequency for the control loop in frames per second.
         display_data: If True, fetches robot observations and displays them in the console and Rerun.
-        display_compressed_images: If True, compresses images before sending them to Rerun for display.
         duration: The maximum duration of the teleoperation loop in seconds. If None, the loop runs indefinitely.
         teleop_action_processor: An optional pipeline to process raw actions from the teleoperator.
         robot_action_processor: An optional pipeline to process actions before they are sent to the robot.
@@ -175,7 +167,6 @@ def teleop_loop(
             log_rerun_data(
                 observation=obs_transition,
                 action=teleop_action,
-                compress_images=display_compressed_images,
             )
 
             print("\n" + "-" * (display_len + 10))
@@ -200,12 +191,7 @@ def teleoperate(cfg: TeleoperateConfig):
     init_logging()
     logging.info(pformat(asdict(cfg)))
     if cfg.display_data:
-        init_rerun(session_name="teleoperation", ip=cfg.display_ip, port=cfg.display_port)
+        init_rerun(session_name="teleoperation")
-    display_compressed_images = (
-        True
-        if (cfg.display_data and cfg.display_ip is not None and cfg.display_port is not None)
-        else cfg.display_compressed_images
-    )
 
     teleop = make_teleoperator_from_config(cfg.teleop)
     robot = make_robot_from_config(cfg.robot)
@@ -224,7 +210,6 @@ def teleoperate(cfg: TeleoperateConfig):
             teleop_action_processor=teleop_action_processor,
             robot_action_processor=robot_action_processor,
             robot_observation_processor=robot_observation_processor,
-            display_compressed_images=display_compressed_images,
         )
     except KeyboardInterrupt:
         pass

src/lerobot/utils/visualization_utils.py

@@ -22,25 +22,13 @@ import rerun as rr
 from .constants import OBS_PREFIX, OBS_STR
 
 
-def init_rerun(
+def init_rerun(session_name: str = "lerobot_control_loop") -> None:
-    session_name: str = "lerobot_control_loop", ip: str | None = None, port: int | None = None
+    """Initializes the Rerun SDK for visualizing the control loop."""
-) -> None:
-    """
-    Initializes the Rerun SDK for visualizing the control loop.
-
-    Args:
-        session_name: Name of the Rerun session.
-        ip: Optional IP for connecting to a Rerun server.
-        port: Optional port for connecting to a Rerun server.
-    """
     batch_size = os.getenv("RERUN_FLUSH_NUM_BYTES", "8000")
     os.environ["RERUN_FLUSH_NUM_BYTES"] = batch_size
     rr.init(session_name)
     memory_limit = os.getenv("LEROBOT_RERUN_MEMORY_LIMIT", "10%")
-    if ip and port:
+    rr.spawn(memory_limit=memory_limit)
-        rr.connect_grpc(url=f"rerun+http://{ip}:{port}/proxy")
-    else:
-        rr.spawn(memory_limit=memory_limit)
 
 
 def _is_scalar(x):
@@ -52,7 +40,6 @@ def _is_scalar(x):
 def log_rerun_data(
     observation: dict[str, Any] | None = None,
     action: dict[str, Any] | None = None,
-    compress_images: bool = False,
 ) -> None:
     """
     Logs observation and action data to Rerun for real-time visualization.
@@ -61,7 +48,7 @@ def log_rerun_data(
     to the Rerun viewer. It handles different data types appropriately:
     - Scalars values (floats, ints) are logged as `rr.Scalars`.
     - 3D NumPy arrays that resemble images (e.g., with 1, 3, or 4 channels first) are transposed
-      from CHW to HWC format, (optionally) compressed to JPEG and logged as `rr.Image` or `rr.EncodedImage`.
+      from CHW to HWC format and logged as `rr.Image`.
     - 1D NumPy arrays are logged as a series of individual scalars, with each element indexed.
     - Other multi-dimensional arrays are flattened and logged as individual scalars.
 
@@ -70,7 +57,6 @@ def log_rerun_data(
     Args:
         observation: An optional dictionary containing observation data to log.
         action: An optional dictionary containing action data to log.
-        compress_images: Whether to compress images before logging to save bandwidth & memory in exchange for cpu and quality.
     """
     if observation:
         for k, v in observation.items():
@@ -89,8 +75,7 @@ def log_rerun_data(
                     for i, vi in enumerate(arr):
                         rr.log(f"{key}_{i}", rr.Scalars(float(vi)))
                 else:
-                    img_entity = rr.Image(arr).compress() if compress_images else rr.Image(arr)
+                    rr.log(key, rr.Image(arr), static=True)
-                    rr.log(key, entity=img_entity, static=True)
 
     if action:
         for k, v in action.items():

tests/utils/test_visualization_utils.py

@@ -41,10 +41,7 @@ def mock_rerun(monkeypatch):
         def __init__(self, arr):
             self.arr = arr
 
-    def dummy_log(key, obj=None, **kwargs):
+    def dummy_log(key, obj, **kwargs):
-        # Accept either positional `obj` or keyword `entity` and record remaining kwargs.
-        if obj is None and "entity" in kwargs:
-            obj = kwargs.pop("entity")
         calls.append((key, obj, kwargs))
 
     dummy_rr = SimpleNamespace(