Dataset Labeling Assignment
Overview
Take-home assignment to test your data labeling skills. You will work with different
types of computer vision labeling tasks including labeling bounding box,
keypoints, and segmentation.
Setup
First, set up a free account in https://www.cvat.ai/ (web version).
Get familiar with the basic features- labeling and exporting. You do not need prior
experience with the tool for this exercise.
Assignment Structure
The assignment has 6 tasks. Each task is in a separate folder in the ZIP file shared
with you.
Your goal is to complete each task and send us back a ZIP file with your answers /
exports from cvat.ai .
Note: cvat.ai has a limit of 1 project per account in the free tier. So to proceed to a
new task/project, you will need to delete your existing project. Make sure you
export the labels before deleting the project.
1. Classification Task
Folder: classification
Task: Sort images into correct categories by placing them in appropriate folders.
What you'll find:
Various images (1.webp, 2.jpg, 3.jpg, etc.)
Two category folders: fire and not-fire
Dataset Labeling Assignment 1
Two category folders: crowded and not-crowded
Instructions:
1. Look at each image carefully
2. Move fire-related images to the fire folder
3. Move non-fire images to the not-fire folder
4. Move crowded scene images to the crowded folder
5. Move non-crowded images to the not-crowded folder
Output: Just the properly sorted images in their respective folders
2. Fire, Smoke & Spark Detection
Folder: fire_smoke_spark_bbox
Task: Draw bounding boxes around fire, smoke, and spark objects.
Classes to detect:
fire
smoke
spark
Instructions:
1. Upload images to CVAT
2. Create a new project or task with the above 3 classes
3. Draw tight bounding boxes around each fire, smoke, or spark object
4. Label each box with the correct class name
5. Export annotations in LabelMe format
Output: Zip file exported from CVAT in LabelMe format
3. Person Detection
Folder: person_bbox
Task: Draw bounding boxes around people in images.
Dataset Labeling Assignment 2
Classes to detect:
person
Instructions:
1. Upload images to CVAT
2. Create a new project or task with "person" class
3. Draw bounding boxes around each person in the images
4. Make sure boxes are tight around the person's body
5. Export annotations in LabelMe format
Output: Zip file exported from CVAT in LabelMe format
4. Object Detection
Folder: objects_bbox
Task: Draw bounding boxes around various objects.
Classes to detect:
tool
car
bus
Instructions:
1. Upload images to CVAT
2. Create a new project or task with "tool", "car" and "bus" classes
3. Draw bounding boxes around each class in the images
4. Label each box with the correct class name
5. Export annotations in LabelMe format
Output: Zip file exported from CVAT in LabelMe format
5. Face Keypoints
Folder: face_keypoints
Dataset Labeling Assignment 3
Task: Mark 5 key points on faces.
Keypoints to mark:
2 eyes (left eye, right eye)
1 nose tip
2 mouth corners (left corner, right corner)
Instructions:
1. Upload images to CVAT
2. Create a new project or task with keypoint annotation type or you can use a
skeleton
3. Set up 5 keypoints as mentioned above
4. Click on the exact location of each keypoint on every face
5. Export annotations in COCO Keypoints format
Output: Zip file exported from CVAT in COCO Keypoints format
6. Segmentation
Folder: segmentation
Task: Create polygon segmentation for objects.
Classes to segment:
car
car windshield
Instructions:
1. Upload images to CVAT
2. Create a new project with polygon annotation type
3. Add the above classes
4. Use polygon tool to trace around objects precisely. Sample:
Dataset Labeling Assignment 4
1. Click multiple points to create accurate object boundaries
2. Make sure polygons are closed and precise
3. Export annotations in LabelMe format
Output: Zip file exported from CVAT in LabelMe format
Submission Requirements
Create a folder with your name and put all files inside it:
Folder structure:
[Your_Full_Name]_labeling_assignment/
├── classification/
│ ├── fire/
│ ├── not-fire/
│ ├── crowded/
│ └── not-crowded/
Dataset Labeling Assignment 5
├── Fire_Smoke_Spark_bbox.zip
├── Person_bbox.zip
├── Objects_bbox.zip
├── Face_keypoints.zip
└── Segmentation.zip
What to submit:
1. Classification folder - with images properly sorted into subfolders
2. Fire_Smoke_Spark_bbox.zip - CVAT export in LabelMe format
3. Person_bbox.zip - CVAT export in LabelMe format
4. Objects_bbox.zip - CVAT export in LabelMe format
5. Face_keypoints.zip - CVAT export in COCO Keypoints format
6. Segmentation.zip - CVAT export in LabelMe format
Example: If your name is "Rahul J H", create a folder named
Rahul_J_H_labeling_assignment and put all the above files inside it.
Dataset Labeling Assignment 6
