3D Segmentation Conversion

A guide for bidirectional V1/V2 conversion of 3D segmentation annotations for PCD (Point Cloud Data).

Data Structure

V1 Structure

# annotations
{
    "id": "seg3d_1",
    "tool": "3d_segmentation",
    "isLocked": False,
    "isVisible": True,
    "isValid": True,
    "classification": {
        "class": "ground",
        "material": "concrete"
    },
    "label": ["ground"]
}

# annotationsData
{
    "id": "seg3d_1",
    "points": [0, 1, 5, 10, 15, 20, 25, 30, 35]
}

V2 Structure

{
    "id": "seg3d_1",
    "classification": "ground",
    "attrs": [
        {"name": "material", "value": "concrete"}
    ],
    "data": {
        "points": [0, 1, 5, 10, 15, 20, 25, 30, 35]
    }
}

Conversion Rules

V1 → V2

V1 Field	V2 Field
points	data.points
classification.class	classification
classification.{other}	attrs[{name, value}]

V2 → V1

V2 Field	V1 Field
data.points	points
classification	classification.class
attrs[{name, value}]	classification.{name: value}

Usage Examples

Basic Conversion

from synapse_sdk.utils.converters.dm import convert_v1_to_v2, convert_v2_to_v1

# V1 3D segmentation data
v1_data = {
    "annotations": {
        "pcd_1": [
            {
                "id": "3DSeg12345",
                "tool": "3d_segmentation",
                "classification": {"class": "ground", "material": "concrete"}
            }
        ]
    },
    "annotationsData": {
        "pcd_1": [
            {
                "id": "3DSeg12345",
                "points": [0, 1, 5, 10, 15, 20, 25, 30, 35]
            }
        ]
    }
}

# Convert to V2
result = convert_v1_to_v2(v1_data)
annotation_data = result["annotation_data"]

# Check V2 result
seg3d = annotation_data["pcds"][0]["3d_segmentation"][0]
print(seg3d["data"]["points"])  # [0, 1, 5, 10, 15, 20, 25, 30, 35]
print(seg3d["classification"])  # "ground"

Large Point Cloud Processing

import numpy as np

# Extract segmentation indices from point cloud
point_cloud = np.random.rand(100000, 3)  # 100,000 points
ground_mask = point_cloud[:, 2] < 0.1  # Select low height points
ground_indices = np.where(ground_mask)[0].tolist()

v1_large = {
    "annotations": {
        "pcd_1": [
            {"id": "GroundSeg", "tool": "3d_segmentation", "classification": {"class": "ground"}}
        ]
    },
    "annotationsData": {
        "pcd_1": [
            {"id": "GroundSeg", "points": ground_indices}
        ]
    }
}

# Convert and verify
result = convert_v1_to_v2(v1_large)
restored = convert_v2_to_v1(result)

original_count = len(v1_large["annotationsData"]["pcd_1"][0]["points"])
restored_count = len(restored["annotationsData"]["pcd_1"][0]["points"])
assert original_count == restored_count

Roundtrip Verification

def verify_3d_seg_roundtrip(v1_original):
    """Verify 3D segmentation roundtrip"""
    # V1 → V2 → V1
    v2_result = convert_v1_to_v2(v1_original)
    v1_restored = convert_v2_to_v1(v2_result)

    # Compare points
    orig_points = v1_original["annotationsData"]["pcd_1"][0]["points"]
    rest_points = v1_restored["annotationsData"]["pcd_1"][0]["points"]

    assert orig_points == rest_points

    print("3D segmentation roundtrip verification successful")

verify_3d_seg_roundtrip(v1_data)

Image vs 3D Segmentation

Property	Image	3D
V1 data field	pixel_indices	points
V2 data type	list[int]	{points: list[int]}
Media ID	image_N	pcd_N
Index meaning	Pixel index	Point index