Support AevaScenes SDK for Python 3.8 and make rerun optional

aevascenes/aevascenes.py

@@ -11,7 +11,6 @@
 
 import cv2
 import numpy as np
-import rerun as rr
 from PIL import Image
 from tqdm import tqdm
 
@@ -21,7 +20,6 @@
 
 import aevascenes.configs as configs
 from aevascenes import utils
-from aevascenes.visualizer import RRVisualizer
 
 
 class AevaScenes:
@@ -42,7 +40,7 @@ def __init__(self, dataroot: str) -> None:
             dataroot: Path to the root directory of the AevaScenes dataset
         """
         self.dataroot = dataroot
-        self.rr_visualizer = RRVisualizer()
+        self.rr_visualizer = None
         metadata_path = os.path.join(self.dataroot, "metadata.json")
         if os.path.exists(metadata_path):
             self.metadata = utils.read_file(metadata_path)
@@ -211,7 +209,13 @@ def visualize_frame(
                 semantic_labels=semantic_labels,
             )
 
-            pcds.append({"lidar_id": lidar_id, "pcd": rr.Points3D(xyz, colors=colors, labels=labels)})
+            pcd_data = {"lidar_id": lidar_id, "xyz": xyz, "colors": colors, "labels": labels}
+            try:
+                import rerun as rr
+                pcd_data["pcd"] = rr.Points3D(xyz, colors=colors, labels=labels)
+            except ImportError:
+                pass
+            pcds.append(pcd_data)
 
         # Load camera metadata
         camera_ids = sequence_metadata["sensors"]["cameras"]
@@ -268,16 +272,25 @@ def visualize_frame(
                 ),
                 interpolation=cv2.INTER_LINEAR,
             )
-            images.append({"camera_id": camera_id, "image": rr.Image(undistorted_image_resized)})
+            img_data = {"camera_id": camera_id, "image_np": undistorted_image_resized}
+            try:
+                import rerun as rr
+                img_data["image"] = rr.Image(undistorted_image_resized)
+            except ImportError:
+                pass
+            images.append(img_data)
 
         # Load object detection boxes
         boxes_serialized = frame["boxes"]
         boxes = utils.deserialize_boxes(boxes_serialized)
-        boxes_rr = utils.convert_boxes_to_rr(boxes, color_map=configs.class_color_map)
-        arrows_rr = utils.convert_box_velocity_arrows_rr(boxes)
 
-        # Send to rerun visualizer
-        self.rr_visualizer.add_data(pcds=pcds, images=images, boxes=boxes_rr, arrows=arrows_rr)
+        # Send to rerun visualizer (if available)
+        if self.rr_visualizer is not None:
+            boxes_rr = utils.convert_boxes_to_rr(boxes, color_map=configs.class_color_map)
+            arrows_rr = utils.convert_box_velocity_arrows_rr(boxes)
+            self.rr_visualizer.add_data(pcds=pcds, images=images, boxes=boxes_rr, arrows=arrows_rr)
+
+        return {"pcds": pcds, "images": images, "boxes": boxes}
 
     def visualize_sampled_frames_from_dataset(
         self, pcd_color_mode: str = "velocity", project_points_on_image: bool = False, image_downsample_factor: int = 1
@@ -303,7 +316,8 @@ def visualize_sampled_frames_from_dataset(
 
             sampled_frame_id = random.randint(0, len(frames) - 1)
 
-            self.rr_visualizer.set_frame_counter(frame_idx=sequence_idx)
+            if self.rr_visualizer is not None:
+                self.rr_visualizer.set_frame_counter(frame_idx=sequence_idx)
             self.visualize_frame(
                 frames[sampled_frame_id],
                 sequence_data["metadata"],
@@ -332,7 +346,8 @@ def visualize_sequence(
         frames = sequence_data["frames"]
 
         for frame_idx in tqdm(range(len(frames)), desc="Visualizing frames"):
-            self.rr_visualizer.set_frame_counter(frame_idx=frame_idx, timestamp_ns=frames[frame_idx]["timestamp_ns"])
+            if self.rr_visualizer is not None:
+                self.rr_visualizer.set_frame_counter(frame_idx=frame_idx, timestamp_ns=frames[frame_idx]["timestamp_ns"])
             self.visualize_frame(
                 frames[frame_idx],
                 sequence_data["metadata"],

aevascenes/configs.py

@@ -50,3 +50,24 @@
     "sidewalk": [85, 0, 100],  # purple
     "other_ground": [150, 100, 0],  # brown
 }
+
+# ── Semantic class groupings ─────────────────────────────────────────
+
+ground_classes = {
+    "road", "lane_boundary", "road_marking", "reflective_marker",
+    "sidewalk", "other_ground",
+}
+
+lane_classes = {
+    "lane_boundary", "road_marking", "reflective_marker",
+}
+
+vehicle_classes = {
+    "car", "bus", "truck", "trailer", "vehicle_on_rails", "other_vehicle",
+}
+
+dynamic_classes = {
+    "car", "bus", "truck", "trailer", "vehicle_on_rails", "other_vehicle",
+    "bicycle", "motorcycle", "motorcyclist", "bicyclist",
+    "pedestrian", "animal",
+}

aevascenes/utils/utils.py

@@ -13,10 +13,8 @@
 import matplotlib.pyplot as plt
 import numpy as np
 import pandas as pd
-import rerun as rr
 import transformations as tf
 import yaml
-from rerun.datatypes import Quaternion
 from scipy.spatial.transform import Rotation as scipy_rot
 
 from aevascenes.utils import colormaps
@@ -260,7 +258,7 @@ def deserialize_boxes(boxes: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
     return boxes
 
 
-def convert_boxes_to_rr(boxes: List[Dict[str, Any]], color_map: Optional[Dict[str, List[int]]] = None) -> rr.Boxes3D:
+def convert_boxes_to_rr(boxes: List[Dict[str, Any]], color_map: Optional[Dict[str, List[int]]] = None):
     """
     Convert bounding boxes to Rerun Boxes3D format for 3D visualization.
     Transforms a list of bounding boxes in AevaScenes format to a Rerun Boxes3D
@@ -280,6 +278,9 @@ def convert_boxes_to_rr(boxes: List[Dict[str, Any]], color_map: Optional[Dict[st
     Returns:
         Rerun Boxes3D object configured for visualization
     """
+    import rerun as rr
+    from rerun.datatypes import Quaternion
+
     boxes_rr = {}
     boxes_rr["half_sizes"] = np.array([box["dimensions"] / 2 for box in boxes])
     boxes_rr["centers"] = np.array([box["center"] for box in boxes])
@@ -305,7 +306,7 @@ def convert_boxes_to_rr(boxes: List[Dict[str, Any]], color_map: Optional[Dict[st
 
 def convert_box_velocity_arrows_rr(
     boxes: List[Dict[str, Any]], color_map: Optional[Dict[str, List[int]]] = None
-) -> rr.Arrows3D:
+):
     """
     Convert bounding box velocity data to Rerun Arrows3D format for motion visualization.
     Creates 3D arrow vectors representing the linear velocity of bounding boxes,
@@ -324,6 +325,8 @@ def convert_box_velocity_arrows_rr(
     Returns:
         Rerun Arrows3D object configured for velocity visualization with:
     """
+    import rerun as rr
+
     arrows_rr = {}
     arrows_rr["centers"] = np.array([box["center"] for box in boxes])
 

aevascenes/visualizer/__init__.py

@@ -4,4 +4,7 @@
 # The AevaScenes dataset is licensed separately under the AevaScenes Dataset License.
 # See https://scenes.aeva.com/license for the full license text.
 
-from .visualizer import RRVisualizer
+try:
+    from .visualizer import RRVisualizer
+except ImportError:
+    RRVisualizer = None

pyproject.toml

@@ -5,7 +5,7 @@ description = "AevaScenes Python SDK"
 authors = [{name = "Aeva, Inc", email = "research@aeva.ai"}]
 license = {file = "LICENSE-CODE"}
 readme = "README.md"
-requires-python = ">=3.10"
+requires-python = ">=3.8"
 dependencies = [
     "tqdm",
     "numpy",
@@ -17,7 +17,6 @@ dependencies = [
     "pillow",
     "scipy",
     "opencv-python",
-    "rerun-sdk==0.24.1",
     "rich",
 ]
 keywords = ["aeva", "aevascenes", "lidar", "fmcw", "sdk"]
@@ -35,12 +34,12 @@ classifiers = [
   "Topic :: Software Development :: Libraries",
 ]
 
-[tool.setuptools]
-packages = ["aevascenes"]
+[tool.setuptools.packages.find]
+where = ["."]
 
 [tool.black]
 line-length = 120
-target-version = ["py310"]
+target-version = ["py38"]
 
 [tool.isort]
 profile = "black"