Fix conversion bug

tier4 · Jan 23, 2025 · 3b30746 · 3b30746
1 parent 0ca57af
commit 3b30746
Show file tree

Hide file tree

Showing 7 changed files with 217 additions and 149 deletions.
diff --git a/config/topic_list_sample.yaml b/config/topic_list_sample.yaml
@@ -8,6 +8,12 @@ mandatory_topic_list:
  - /sensing/camera/camera3/image_rect_color/compressed
  - /sensing/camera/camera4/image_rect_color/compressed
  - /sensing/camera/camera5/image_rect_color/compressed
+#  - /sensing/camera/camera0/image_raw/compressed
+#  - /sensing/camera/camera1/image_raw/compressed
+#  - /sensing/camera/camera2/image_raw/compressed
+#  - /sensing/camera/camera3/image_raw/compressed
+#  - /sensing/camera/camera4/image_raw/compressed
+#  - /sensing/camera/camera5/image_raw/compressed
  - /localization/kinematic_state
 topic_list:
  - /sensing/lidar/left/velodyne_packets
@@ -21,6 +27,12 @@ topic_list:
  - /sensing/camera/camera3/image_rect_color/compressed
  - /sensing/camera/camera4/image_rect_color/compressed
  - /sensing/camera/camera5/image_rect_color/compressed
+#  - /sensing/camera/camera0/image_raw/compressed
+#  - /sensing/camera/camera1/image_raw/compressed
+#  - /sensing/camera/camera2/image_raw/compressed
+#  - /sensing/camera/camera3/image_raw/compressed
+#  - /sensing/camera/camera4/image_raw/compressed
+#  - /sensing/camera/camera5/image_raw/compressed
  - /sensing/camera/camera0/camera_info
  - /sensing/camera/camera1/camera_info
  - /sensing/camera/camera2/camera_info

diff --git a/perception_dataset/constants.py b/perception_dataset/constants.py
@@ -6,7 +6,7 @@
 class T4_FORMAT_DIRECTORY_NAME(Enum):
     ANNOTATION = "annotation"
     DATA = "data"
-    LIDARSEG_ANNO_FOLDER = "lidar_seg"
+    LIDARSEG_ANNO_FOLDER = "lidarseg"
 
 
 class SENSOR_MODALITY_ENUM(Enum):

diff --git a/perception_dataset/deepen/deepen_to_t4_converter.py b/perception_dataset/deepen/deepen_to_t4_converter.py
@@ -97,7 +97,7 @@ def _format_annotations(self) -> Dict[str, Dict[str, Any]]:
         )
         return scenes_anno_dict
 
-    def _format_point_3d_annotations(self) -> Dict[str, Dict[Any]]:
+    def _format_point_3d_annotations(self) -> Dict[str, Dict[str, Any]]:
         """Format annotations for point_3d from input_anno_file."""
         if self._label_info.label_format == LabelFormat.PAINT_3D:
             deepen_annotations = DeepenSegmentationPainting3DAnnotations.from_file(

diff --git a/perception_dataset/deepen/download_annotations.py b/perception_dataset/deepen/download_annotations.py
@@ -14,7 +14,6 @@
 
 CLIENT_ID = os.environ["DEEPEN_CLIENT_ID"]
 ACCESS_TOKEN = os.environ["DEEPEN_ACCESS_TOKEN"]
-DATASET_ID = "r82dS6VVkok98ZWr1CTkU4NE"
 _HEADERS = {
     "Authorization": f"Bearer {ACCESS_TOKEN}",
     "Content-Type": "application/json",
@@ -46,69 +45,90 @@ def get_paint3d_labels(dataset_ids: List[str], dataset_dir: str, output_name: st
     """
     Get 3d paint labels to lidar_seg/<dataset_id>_<frame_id>.bin and <output_name>.json.
     Example of a response:
-        {
-            'server': 'gunicorn', 'date': 'Wed, 15 Jan 2025 08:45:01 GMT',
-            'content-type': 'text/html; charset=utf-8',
-            'paint-metadata': '{
-                "format": "pako_compressed", "paint_categories": ["car", "wall"],
-                "frame_sizes": [173430, 173097, 172595, 172957, 173011, 173204, 173813, 173913, 174393, 175304, 175176, 174038, 174473, 174373, 172598,
-                171886, 172377, 171724, 173322, 173566, 175911, 176165, 177180, 177774, 178544, 179606, 178347,
-                177048, 176775, 176997, 177271, 177864, 179341, 178986, 179670, 179973, 178885, 178424, 179233, 179766, 178644, 179057, 177871, 178286, 178570,
-                177410, 176688, 176961, 176871, 176862, 176740, 177872, 179313, 179414, 179496, 178974, 178653, 178992, 178861, 178823, 178690, 178815, 178966, 178960, 178907]
-            }',
-            'vary': 'Accept, Cookie, Accept-Encoding',
-            'allow': 'POST, GET, OPTIONS',
-            'x-frame-options': 'DENY',
-            'x-content-type-options': 'nosniff',
-            'referrer-policy': 'same-origin',
-            'content-encoding': 'gzip',
-            'via': '1.1 google',
-            'Strict-Transport-Security': 'max-age=31536000; includeSubDomains',
-            'Alt-Svc': 'h3=":443"; ma=2592000,h3-29=":443"; ma=2592000',
-            'Transfer-Encoding': 'chunked'
-        }
+            {
+                    'server': 'gunicorn', 'date': 'Wed, 15 Jan 2025 08:45:01 GMT',
+                    'content-type': 'text/html; charset=utf-8',
+                    'paint-metadata': '{
+                            "format": "pako_compressed", "paint_categories": ["car", "wall"],
+                            "frame_sizes": [173430, 173097, 172595, 172957, 173011, 173204, 173813, 173913, 174393, 175304, 175176, 174038, 174473, 174373, 172598,
+                            171886, 172377, 171724, 173322, 173566, 175911, 176165, 177180, 177774, 178544, 179606, 178347,
+                            177048, 176775, 176997, 177271, 177864, 179341, 178986, 179670, 179973, 178885, 178424, 179233, 179766, 178644, 179057, 177871, 178286, 178570,
+                            177410, 176688, 176961, 176871, 176862, 176740, 177872, 179313, 179414, 179496, 178974, 178653, 178992, 178861, 178823, 178690, 178815, 178966, 178960, 178907]
+                    }',
+                    'vary': 'Accept, Cookie, Accept-Encoding',
+                    'allow': 'POST, GET, OPTIONS',
+                    'x-frame-options': 'DENY',
+                    'x-content-type-options': 'nosniff',
+                    'referrer-policy': 'same-origin',
+                    'content-encoding': 'gzip',
+                    'via': '1.1 google',
+                    'Strict-Transport-Security': 'max-age=31536000; includeSubDomains',
+                    'Alt-Svc': 'h3=":443"; ma=2592000,h3-29=":443"; ma=2592000',
+                    'Transfer-Encoding': 'chunked'
+            }
 
     <output_name>.json is in the following format:
-        [
-            {
-                "dataset_id": "DOnC2vK05ojPr7qiqCsk2Ee7",
-                "file_id": "0.pcd",
-                "label_type": "3d_point",
-                "label_id": "none",		# Keep it for consistency with downstream tasks
-                "label_category_id": "none",	# Keep it for consistency with downstream tasks
-                "total_lidar_points": 173430,
-                        "sensor_id": "lidar",
-                        "stage_id": "QA",
-                "paint_categories": ["car", "wall", ...],
-                "lidarseg_anno_file": "lidar_seg/DOnC2vK05ojPr7qiqCsk2Ee7_0.bin"
-            },
-            ...
-        ]
+            [
+                    {
+                            "dataset_id": "DOnC2vK05ojPr7qiqCsk2Ee7",
+                            "file_id": "0.pcd",
+                            "label_type": "3d_point",
+                            "label_id": "none",		# Keep it for consistency with downstream tasks
+                            "label_category_id": "none",	# Keep it for consistency with downstream tasks
+                            "total_lidar_points": 173430,
+                                            "sensor_id": "lidar",
+                                            "stage_id": "QA",
+                            "paint_categories": ["car", "wall", ...],
+                            "lidarseg_anno_file": "lidar_seg/DOnC2vK05ojPr7qiqCsk2Ee7_0.bin"
+                    },
+                    ...
+            ]
     """
-    dataset_path = Path(dataset_dir) / T4_FORMAT_DIRECTORY_NAME.LIDARSEG_ANNO_FOLDER.value()
+    dataset_path = Path(dataset_dir) / T4_FORMAT_DIRECTORY_NAME.LIDARSEG_ANNO_FOLDER.value
     dataset_path.mkdir(parents=True, exist_ok=True)
     lidarseg_annos_info = []
     for dataset_id in dataset_ids:
-        datasets_url = f"https://tools.deepen.ai/api/v2/datasets/{DATASET_ID}/label_types/3d_point/paint_labels?stageId=QA&sensor_id=lidar&labelSetId=default"
+        datasets_url = f"https://tools.deepen.ai/api/v2/datasets/{dataset_id}/label_types/3d_point/paint_labels?stageId=QA&sensor_id=lidar&labelSetId=default"
         try:
             response = requests.get(datasets_url, headers=_HEADERS)
             response.raise_for_status()
         except requests.exceptions.RequestException as e:
             raise SystemExit(e)
 
+        # Get pcd files
+        label_url = f"https://tools.deepen.ai/api/v2/datasets/{dataset_id}/labels?labels_with_file_ids_order=true&final=true"
+        try:
+            label_response = requests.get(label_url, headers=_HEADERS)
+            label_response.raise_for_status()
+        except requests.exceptions.RequestException as e:
+            raise SystemExit(e)
+
+        label_response_data = label_response.json()
+        pcd_files = [
+            pcd_file_name
+            for pcd_file_name in label_response_data["labels"].keys()
+            if pcd_file_name.endswith(EXTENSION_ENUM.PCD.value)
+        ]
+
         paint_metadata = json.loads(response.headers["paint-metadata"])
+        assert len(pcd_files) == len(
+            paint_metadata["frame_sizes"]
+        ), "Number of pcd files and frame sizes do not match!"
+
         decompressed_data = zlib.decompress(response.content)
         lidarseg_category_indices = np.frombuffer(decompressed_data, dtype=np.uint8)
         # Split based on pointcloud size
         previous_frame_size = 0
-        for file_id, frame_size in enumerate(paint_metadata["frame_sizes"]):
+
+        for file_id, frame_size in zip(pcd_files, paint_metadata["frame_sizes"]):
+
             lidarseg_anno_filename = (
                 dataset_path / f"{dataset_id}_{str(file_id)}{EXTENSION_ENUM.BIN.value}"
             )
 
             annos_info = {
                 "dataset_id": dataset_id,
-                "file_id": f"{file_id}.pcd",
+                "file_id": file_id,
                 "label_id": "none",  # Keep it for consistency with downstream tasks
                 "label_category_id": "none",  # Keep it for consistency with downstream tasks
                 "total_lidar_points": frame_size,
@@ -127,11 +147,12 @@ def get_paint3d_labels(dataset_ids: List[str], dataset_dir: str, output_name: st
             lidarseg_annos_info.append(annos_info)
 
     # Save lidarseg anno information to json
-    output_file = os.path.join(dataset_dir, output_name)
-    with open(output_file, "w") as f:
+    output_name = Path(dataset_dir, output_name)
+    output_name.parents[0].mkdir(parents=True, exist_ok=True)
+    with open(output_name, "w") as f:
         json.dump(lidarseg_annos_info, f, indent=4)
 
-    print(f"Annotation file is saved: {output_file}")
+    print(f"Annotation file is saved: {output_name}")
 
 
 def get_datasets(dataset_ids: List[str], dataset_dir: str, output_name: str, label_type: str):

diff --git a/perception_dataset/deepen/segmentation/painting3d.py b/perception_dataset/deepen/segmentation/painting3d.py
@@ -1,7 +1,8 @@
 from __future__ import annotations
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 import json
+import re
 from typing import Any, Dict, List
 
 from perception_dataset.deepen.deepen_annotation import DeepenAnnotation, LabelType
@@ -22,9 +23,9 @@ class DeepenSegmentationPainting3DScene(DeepenAnnotation):
     :param total_lidar_points: Total number of lidar pointclouds.
     """
 
-    paint_categories: List[str]
-    lidarseg_anno_file: str
-    total_lidar_points: int
+    paint_categories: List[str] = field(default_factory=list)
+    lidarseg_anno_file: str = ""
+    total_lidar_points: int = 0
 
     def __post_init__(self) -> None:
         """
@@ -33,6 +34,7 @@ def __post_init__(self) -> None:
         """
         assert len(self.paint_categories) > 0, "Length of categories must be more than 0!"
         assert self.total_lidar_points > 0, "Lidar pointclouds must be more than 0!"
+        assert self.lidarseg_anno_file != "", "Must provide a lidarseg annotation file!"
 
 
 @dataclass
@@ -48,21 +50,22 @@ def add_record_to_dataset(
         self, segmentation_painting_3d_scene: DeepenSegmentationPainting3DScene
     ) -> None:
         """Add a DeepenSegmentationPainting3DScene to the dataset records."""
-        scene_id = segmentation_painting_3d_scene.file_id
+        filename = segmentation_painting_3d_scene.file_id.split(".")[0]
+        scene_id = int(re.sub(r"\D", "", filename[-6:]))
         self.deepen_segmentation_scenes[scene_id] = segmentation_painting_3d_scene
 
     def format_scene_annotations(self) -> Dict[int, List[Dict[str, Any]]]:
         """
         Format scene annotations to
         {
-            id:
-            [
-                {
-                    "paint_categories": ["car", "wall", ...],
-                    "lidarseg_anno_file": "lidar_seg/DOnC2vK05ojPr7qiqCsk2Ee7_0.bin",
-                    ...
-                }
-            ]
+                id:
+                [
+                        {
+                                "paint_categories": ["car", "wall", ...],
+                                "lidarseg_anno_file": "lidar_seg/DOnC2vK05ojPr7qiqCsk2Ee7_0.bin",
+                                ...
+                        }
+                ]
         }
         """
         return {
@@ -87,30 +90,30 @@ def from_file(
         """Return DeepenSegmentationPainting3DDataset from files.
 
         Args:
-            ann_file (str): Annotation files path in json. The format is:
-            [
-                {
-                    "dataset_id": "DOnC2vK05ojPr7qiqCsk2Ee7",
-                    "file_id": "0.pcd",
-                    "label_type": "3d_point",
-                    "label_id": "none",		# Keep it for consistency with downstream tasks
-                    "label_category_id": "none",	# Keep it for consistency with downstream tasks
-                    "total_lidar_points": 173430,
-                    "sensor_id": "lidar",
-                    "stage_id": "QA",
-                    "paint_categories": ["car", "wall", ...],
-                    "lidarseg_anno_file": "lidar_seg/DOnC2vK05ojPr7qiqCsk2Ee7_0.bin"
-                },
-                ...
-            ]
-            data_root (str): Root directory of the T4 dataset.
-            camera2index (Dict[str, int]): Name mapping from camera name to camera index.
-            dataset_corresponding (Dict[str, str]): Key-value mapping of T4 dataset name and Deepen ID.
-            as_dict (bool, optional): Whether to output objects as dict or its instance.
-                Defaults to True.
+                ann_file (str): Annotation files path in json. The format is:
+                [
+                        {
+                                "dataset_id": "DOnC2vK05ojPr7qiqCsk2Ee7",
+                                "file_id": "0.pcd",
+                                "label_type": "3d_point",
+                                "label_id": "none",		# Keep it for consistency with downstream tasks
+                                "label_category_id": "none",	# Keep it for consistency with downstream tasks
+                                "total_lidar_points": 173430,
+                                "sensor_id": "lidar",
+                                "stage_id": "QA",
+                                "paint_categories": ["car", "wall", ...],
+                                "lidarseg_anno_file": "lidar_seg/DOnC2vK05ojPr7qiqCsk2Ee7_0.bin"
+                        },
+                        ...
+                ]
+                data_root (str): Root directory of the T4 dataset.
+                camera2index (Dict[str, int]): Name mapping from camera name to camera index.
+                dataset_corresponding (Dict[str, str]): Key-value mapping of T4 dataset name and Deepen ID.
+                as_dict (bool, optional): Whether to output objects as dict or its instance.
+                        Defaults to True.
 
         Returns:
-            List[DeepenSegmentationPainting2D]: List of converted `DeepenSegmentationPainting2D`s.
+                List[DeepenSegmentationPainting2D]: List of converted `DeepenSegmentationPainting2D`s.
         """
         with open(ann_file, "r") as f:
             lidarseg_ann_info = json.load(f)
@@ -146,15 +149,15 @@ def from_file(
     def format_deepen_annotations(self) -> Dict[str, Dict[int, List[Dict[str, Any]]]]:
         """
         Convert to {
-            dataset_id: {
-                scene_id/frame_index: [
-                    {
-                        "paint_categories": ["car", "wall", ...],
-                        "lidarseg_anno_file": "lidar_seg/DOnC2vK05ojPr7qiqCsk2Ee7_0.bin",
-                        ...
-                    }
-                ]
-            }
+                dataset_id: {
+                        scene_id/frame_index: [
+                                {
+                                        "paint_categories": ["car", "wall", ...],
+                                        "lidarseg_anno_file": "lidar_seg/DOnC2vK05ojPr7qiqCsk2Ee7_0.bin",
+                                        ...
+                                }
+                        ]
+                }
         }
         """
         return {