annotation import

CHANGELOG.md

@@ -2,7 +2,10 @@
 
 # Next Release
 ## Added
-* Added new `Model` schema
+* Added new `Model` and 'ModelRun` entities
+* Update client to support creating and querying for `Model`s
+* Implement new prediction import pipeline to support both MAL and MEA
+* Added notebook to demonstrate how to use MEA
 
 # Version 2.5.6 (2021-05-19)
 ## Fix

examples/model_assisted_labeling/ndjson_utils.py

@@ -0,0 +1,126 @@
+from labelbox import Client
+
+from typing import Dict, Any, Tuple
+from skimage import measure
+from io import BytesIO
+from PIL import Image
+import numpy as np
+import uuid
+
+
+def create_boxes_ndjson(datarow_id: str, schema_id: str, top: float, left: float,
+                        bottom: float, right: float) -> Dict[str, Any]:
+    """
+    * https://docs.labelbox.com/data-model/en/index-en#bounding-box
+
+    Args:
+        datarow_id (str): id of the data_row (in this case image) to add this annotation to
+        schema_id (str): id of the bbox tool in the current ontology
+        top, left, bottom, right (int): pixel coordinates of the bbox
+    Returns:
+        ndjson representation of a bounding box
+    """
+    return {
+        "uuid": str(uuid.uuid4()),
+        "schemaId": schema_id,
+        "dataRow": {
+            "id": datarow_id
+        },
+        "bbox": {
+            "top": int(top),
+            "left": int(left),
+            "height": int(bottom - top),
+            "width": int(right - left)
+        }
+    }
+
+
+def create_polygon_ndjson(datarow_id: str, schema_id: str,
+                          segmentation_mask: np.ndarray) -> Dict[str, Any]:
+    """
+    * https://docs.labelbox.com/data-model/en/index-en#polygon
+
+    Args:
+        datarow_id (str): id of the data_row (in this case image) to add this annotation to
+        schema_id (str): id of the bbox tool in the current ontology
+        segmentation_mask (np.ndarray): Segmentation mask of size (image_h, image_w)
+            - Seg mask is turned into a polygon since polygons aren't directly inferred.
+    Returns:
+        ndjson representation of a polygon
+    """
+    contours = measure.find_contours(segmentation_mask, 0.5)
+    #Note that complex polygons could break.
+    pts = contours[0].astype(np.int32)
+    pts = np.roll(pts, 1, axis=-1)
+    pts = [{'x': int(x), 'y': int(y)} for x, y in pts]
+    return {
+        "uuid": str(uuid.uuid4()),
+        "schemaId": schema_id,
+        "dataRow": {
+            "id": datarow_id
+        },
+        "polygon": pts
+    }
+
+
+def create_mask_ndjson(client: Client, datarow_id: str, schema_id: str,
+                       segmentation_mask: np.ndarray, color: Tuple[int, int,
+                                                     int]) -> Dict[str, Any]:
+    """
+    Creates a mask for each object in the image
+    * https://docs.labelbox.com/data-model/en/index-en#segmentation-mask
+
+    Args:
+        client (labelbox.Client): labelbox client used for uploading seg mask to google cloud storage
+        datarow_id (str): id of the data_row (in this case image) to add this annotation to
+        schema_id (str): id of the segmentation tool in the current ontology
+        segmentation_mask is a segmentation mask of size (image_h, image_w)
+        color ( Tuple[int,int,int]): rgb color to convert binary mask into 3D colorized mask
+    Return:
+        ndjson representation of a segmentation mask
+    """
+
+    colorize = np.concatenate(([segmentation_mask[..., np.newaxis] * c for c in color]),
+                              axis=2)
+    img_bytes = BytesIO()
+    Image.fromarray(colorize).save(img_bytes, format="PNG")
+    #* Use your own signed urls so that you can resign the data
+    #* This is just to make the demo work
+    url = client.upload_data(content=img_bytes.getvalue(), sign=True)
+    return {
+        "uuid": str(uuid.uuid4()),
+        "schemaId": schema_id,
+        "dataRow": {
+            "id": datarow_id
+        },
+        "mask": {
+            "instanceURI": url,
+            "colorRGB": color
+        }
+    }
+
+
+def create_point_ndjson(datarow_id: str, schema_id: str, top: float, left: float,
+                        bottom: float, right: float) -> Dict[str, Any]:
+    """
+    * https://docs.labelbox.com/data-model/en/index-en#point
+
+    Args:
+        datarow_id (str): id of the data_row (in this case image) to add this annotation to
+        schema_id (str): id of the point tool in the current ontology
+        t, l, b, r (int): top, left, bottom, right pixel coordinates of the bbox
+        - The model doesn't directly predict points, so we grab the centroid of the predicted bounding box
+    Returns:
+        ndjson representation of a polygon
+    """
+    return {
+        "uuid": str(uuid.uuid4()),
+        "schemaId": schema_id,
+        "dataRow": {
+            "id": datarow_id
+        },
+        "point": {
+            "x": int((left + right) / 2.),
+            "y": int((top + bottom) / 2.),
+        }
+    }

labelbox/__init__.py

@@ -2,7 +2,9 @@
 __version__ = "2.5.6"
 
 from labelbox.client import Client
+from labelbox.schema.model import Model
 from labelbox.schema.bulk_import_request import BulkImportRequest
+from labelbox.schema.annotation_import import MALPredictionImport, MEAPredictionImport
 from labelbox.schema.project import Project
 from labelbox.schema.dataset import Dataset
 from labelbox.schema.data_row import DataRow
@@ -19,4 +21,3 @@
 from labelbox.schema.role import Role, ProjectRole
 from labelbox.schema.invite import Invite, InviteLimit
 from labelbox.schema.model_run import ModelRun
-from labelbox.schema.model import Model

labelbox/client.py

@@ -79,7 +79,13 @@ def __init__(self,
 
     @retry.Retry(predicate=retry.if_exception_type(
         labelbox.exceptions.InternalServerError))
-    def execute(self, query, params=None, timeout=30.0, experimental=False):
+    def execute(self,
+                query=None,
+                params=None,
+                data=None,
+                files=None,
+                timeout=30.0,
+                experimental=False):
         """ Sends a request to the server for the execution of the
         given query.
 
@@ -89,6 +95,8 @@ def execute(self, query, params=None, timeout=30.0, experimental=False):
         Args:
             query (str): The query to execute.
             params (dict): Query parameters referenced within the query.
+            data (str): json string containing the query to execute
+            files (dict): file arguments for request
             timeout (float): Max allowed time for query execution,
                 in seconds.
         Returns:
@@ -107,8 +115,9 @@ def execute(self, query, params=None, timeout=30.0, experimental=False):
                 most likely due to connection issues.
             labelbox.exceptions.LabelboxError: If an unknown error of any
                 kind occurred.
+            ValueError: If query and data are both None.
         """
-        logger.debug("Query: %s, params: %r", query, params)
+        logger.debug("Query: %s, params: %r, data %r", query, params, data)
 
         # Convert datetimes to UTC strings.
         def convert_value(value):
@@ -117,19 +126,35 @@ def convert_value(value):
                 value = value.strftime("%Y-%m-%dT%H:%M:%SZ")
             return value
 
-        if params is not None:
-            params = {
-                key: convert_value(value) for key, value in params.items()
-            }
-
-        data = json.dumps({'query': query, 'variables': params}).encode('utf-8')
-
+        if query is not None:
+            if params is not None:
+                params = {
+                    key: convert_value(value) for key, value in params.items()
+                }
+            data = json.dumps({
+                'query': query,
+                'variables': params
+            }).encode('utf-8')
+        elif data is None:
+            raise ValueError("query and data cannot both be none")
         try:
-            response = requests.post(self.endpoint.replace('/graphql', '/_gql')
-                                     if experimental else self.endpoint,
-                                     data=data,
-                                     headers=self.headers,
-                                     timeout=timeout)
+            request = {
+                'url':
+                    self.endpoint.replace('/graphql', '/_gql')
+                    if experimental else self.endpoint,
+                'data':
+                    data,
+                'headers':
+                    self.headers,
+                'timeout':
+                    timeout
+            }
+            if files:
+                request.update({'files': files})
+                request['headers'] = {
+                    'Authorization': self.headers['Authorization']
+                }
+            response = requests.post(**request)
             logger.debug("Response: %s", response.text)
         except requests.exceptions.Timeout as e:
             raise labelbox.exceptions.TimeoutError(str(e))
@@ -548,4 +573,14 @@ def create_model(self, name, ontology_id):
             InvalidAttributeError: If the Model type does not contain
                 any of the attribute names given in kwargs.
         """
-        return self._create(Model, {"name": name, "ontology_id": ontology_id})
+        query_str = """mutation createModelPyApi($name: String!, $ontologyId: ID!){
+            createModel(data: {name : $name, ontologyId : $ontologyId}){
+                    %s
+                }
+            }""" % query.results_query_part(Model)
+
+        result = self.execute(query_str, {
+            "name": name,
+            "ontologyId": ontology_id
+        })
+        return Model(self, result['createModel'])

labelbox/orm/db_object.py

@@ -48,7 +48,6 @@ def __init__(self, client, field_values):
             if relationship.cache and value is None:
                 raise KeyError(
                     f"Expected field  values for {relationship.name}")
-
             setattr(self, relationship.name,
                     RelationshipManager(self, relationship, value))
 

labelbox/schema/__init__.py

@@ -1,5 +1,6 @@
 import labelbox.schema.asset_metadata
 import labelbox.schema.bulk_import_request
+import labelbox.schema.annotation_import
 import labelbox.schema.benchmark
 import labelbox.schema.data_row
 import labelbox.schema.dataset