Skip to content
intermediateimage

How to annotate Object Detection

Object detection is the task of finding each object of interest in an image and marking it with a bounding box plus a class label.

Draw bounding boxes around objects for object detection model training.

Get This Design

View on GitHub(opens in a new tab)

Clone or download from the repository

Quick start:

bash
git clone https://github.com/davidjurgens/potato-showcase.git
cd potato-showcase/templates/image/object-detection
potato start config.yaml

About this dataset

Object detection is the task of finding each object of interest in an image and marking it with a bounding box plus a class label. It is one of the core supervised computer-vision tasks and supplies the training data that detectors such as the YOLO and Faster R-CNN families learn from.

This entry is a generic annotation template rather than a specific published dataset. The images are placeholders, and the label set is a small illustrative one (person, car, dog, cat) meant to be swapped for whatever classes a project needs. The annotation task is to draw a tight box around every instance of a target class and assign it a label.

Because it is a template, there is no fixed corpus, category count, or benchmark split to report. In practice, bounding-box detection is defined by public datasets like COCO and PASCAL VOC, which are the canonical references for evaluation and label conventions but are not reproduced here.

The Potato config below reproduces this task with a single image_annotation scheme using the bbox tool and a short list of class labels, writing each box as coordinates in JSON output. It is meant as a starting point that you point at your own images and edit the label set to build a detection training set.

Task type
Object detection
Annotation method
Bounding boxes
Potato tool
image_annotation (bbox)
Example labels
Person, Car, Dog, Cat
Output format
JSON box coordinates
Canonical references
COCO, PASCAL VOC

Configuration File

This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.

yaml
annotation_task_name: "Object Detection"
task_name: "Object Detection"
task_description: "Draw bounding boxes around all objects of interest."
task_dir: "."
port: 8000

data_files:
  - "sample-data.json"

item_properties:
  id_key: "id"
  text_key: "image_url"
  image_key: image_url

annotation_schemes:
  - annotation_type: image_annotation
    name: objects
    description: "Draw boxes around objects"
    tools:
      - bbox
    labels:
      - "Person"
      - "Car"
      - "Dog"
      - "Cat"

output_annotation_dir: "output/"
output_annotation_format: "json"

Sample Data

json
[
  {
    "id": "1",
    "image_url": "https://images.unsplash.com/photo-1517849845537-4d257902454a?w=640"
  },
  {
    "id": "2",
    "image_url": "https://images.unsplash.com/photo-1506905925346-21bda4d32df4?w=640"
  }
]

Details

Annotation Types

image_annotation

Domain

Computer Vision

Use Cases

Object DetectionBounding Box

Tags

object-detectionbboxcomputer-visionyolo

Found an issue or want to improve this design?

Open an Issue(opens in a new tab)

Related Designs

Image Segmentation

Draw polygon masks around objects for semantic segmentation tasks.

image_annotation

PixMo-Points - Pointing Annotation for Vision-Language Models

Collect the pointing supervision behind Molmo: place a point on what each referring expression refers to. Scored as a point-in-region hit rate, because a point has no area and every IoU against one is zero.

grounding_evalimage_annotation

RefCOCO - Referring Expression Grounding

Visual grounding task where annotators bind each referring expression to a region of the image. Based on RefCOCO (Yu et al., ECCV 2016), reproduced with Potato's grounding_eval schema, which scores grounding by IoU at four thresholds and counts ungroundedness separately.

grounding_evalimage_annotation