Comment on page
Objects binding
This guide explains how to bind (group) objects on images
For some labeling scenarios, it is needed to group objects. Let's consider the case when you need to label object parts and then group them together. Such binding will help you distinguish parts of different objects. In this tutorial, you will learn how to programmatically group objects together (binding) and how to work with existing bindings.
In this tutorial, we will create binding for the labeled parts of a single car:

Imput image and masks
This tutorial consists of two parts:
git clone https://github.com/supervisely-ecosystem/tutorial-object-binding
cd tutorial-object-binding
./create_venv.sh
Step 3. Open repository directory in Visual Studio Code.
code -r .
Step 4. change ✅ workspace ID ✅ in
local.env
file by copying the ID from the context menu of the workspace. A new project with demo data will be created in the workspace you define:WORKSPACE_ID=619 # ⬅️ change value

Copy workspace ID from context menu
Step 5. Start debugging
src/main.py
import os
from typing import List
from dotenv import load_dotenv
import cv2
import uuid
import supervisely as sly
Init API for communicating with Supervisely Instance. First, we load environment variables with credentials and workspace ID:
if sly.is_development():
load_dotenv("local.env")
load_dotenv(os.path.expanduser("~/supervisely.env"))
api = sly.Api()
Create empty project with name "tutorial-bindings" with one dataset "dataset-01" in your workspace on server. If the project with the same name exists in your dataset, it will be automatically renamed (tutorial-bindings_001, tutorial-bindings_002, etc ...) to avoid name collisions.
workspace_id = sly.env.workspace_id()
project = api.project.create(workspace_id, name="tutorial-bindings", change_name_if_conflict=True)
dataset = api.dataset.create(project.id, name="dataset-01")
print(f"Open project in browser: {project.url}")
Let's define annotation classes and upload them to our new project on server:
class_car = sly.ObjClass(name="car", geometry_type=sly.Bitmap, color=[255, 0, 255])
meta = sly.ProjectMeta(obj_classes=[class_car])
api.project.update_meta(project.id, meta)
image_path = "images/image.jpg"
image_name = sly.fs.get_file_name_with_ext(image_path)
image_info = api.image.upload_path(dataset.id, image_name, image_path)
print(f"Image has been successfully uploaded: id={image_info.id}")
# will be needed later for creating annotation
height, width = cv2.imread(image_path).shape[0:2]
# create Supervisely annotation from masks images
labels_masks: List[sly.Label] = []
for mask_path in [
"images/car_masks/car_01.png",
"images/car_masks/car_02.png",
"images/car_masks/car_03.png",
]:
# read only first channel of the black-and-white image
image_black_and_white = cv2.imread(mask_path)[:, :, 0]
# supports masks with values (0, 1) or (0, 255) or (False, True)
mask = sly.Bitmap(image_black_and_white)
label = sly.Label(geometry=mask, obj_class=class_car)
labels_masks.append(label)
ann = sly.Annotation(img_size=[height, width], labels=labels_masks)
We know that all three masks are parts of a single car object. Let's bind them together. It is important to notice that any unique string can be label's
binding_key
.key = uuid.uuid4().hex # key can be any unique string
for label in ann.labels:
label.binding_key = key
api.annotation.upload_ann(image_info.id, ann)
As a result, we will have three objects of class car grouped together:

Result annotation with binding
project_meta = sly.ProjectMeta.from_json(api.project.get_meta(project.id))
ann_json = api.annotation.download_json(image_info.id)
ann = sly.Annotation.from_json(ann_json, meta)
If
binding_key
is None then the label does not belong to any group.print("Labels bindings:")
for label in ann.labels:
print(label.binding_key)
The output will be the following:
Labels bindings:
62245779
62245779
62245779
groups = ann.get_bindings()
for i, (binding_key, labels) in enumerate(groups.items()):
if binding_key is not None:
print(f"Group # {i} [key={binding_key}] has {len(labels)} labels")
else:
# Binding key None defines all labels that do not belong to any binding group
print(f"{len(labels)} labels do not have binding")
Let's remove bindings on objects of class car:
for label in ann.labels:
if label.obj_class.name == "car":
label.binding_key = None
Or you can discard binding for all labels in annotation:
ann.discard_bindings()
Let's upload updated annotation (without bindings) back to the server:
api.annotation.upload_ann(image_info.id, ann)
Last modified 8mo ago