Posted: Jun 3, 2026
Wall and Floor instance segmentation
CVATImageSegmentationHindiPay Per LabelProject Overview
The goal of this project is to create a high-quality, pixel-accurate instance segmentation dataset of indoor residential and commercial spaces. Annotators will identify and trace the exact boundaries of individual walls and floors. Unlike semantic segmentation, this is an instance-level task. If a room has multiple visible walls (e.g., a left wall, a back wall, and a right wall), they must be annotated as separate, individual instances. 2. Target Classes & Annotation Rules Class 1: wall (Instance-level) Definition: Any vertical wall plane separating rooms or defining the indoor perimeter. Separation Rule: Each individual wall plane must be annotated as a separate instance. Walls are separated by: 3D corners (where two wall planes meet at an angle). Structural pillars, columns, or major wall protrusions. Doorways, archways, or room transitions. Occlusion Rule: Do NOT label doors, windows, paintings, shelves, or furniture as part of the wall. Truncate the polygon tightly around these foreground objects. Class 2: floor (Instance-level) Definition: The walking surface or ground plane of the room. Separation Rule: In open-plan spaces (e.g., kitchen-living room transitions), separate floors into individual instances if there is a change in flooring material (e.g., tile to hardwood) or a clear structural threshold. Occlusion Rule: Exclude rugs, carpets (unless it is wall-to-wall carpeting), furniture bases, and other floor-mounted objects from the floor polygon. 3. Detailed Annotation Guidelines Pixel Accuracy: Polygons must be drawn tightly along boundaries. A maximum deviation of ± 2-3 pixels from the true boundary is acceptable. Foreground Occlusions: If an object (like a chair or plant) stands in front of a wall, the wall polygon must trace around the object. If the object splits a single wall plane into two visible parts, annotate them as two separate polygons but group them under the same instance ID (if the platform supports grouping) or annotate them as separate polygons. Pillars and Columns: Treat structural pillars/columns as separate wall instances. Ceiling & Skirting Boards: Skirting boards (baseboards) should be included as part of the wall instance. Ceilings are excluded from this project. 4. Quality Control & Instructions for Annotators Zoom Level: Annotators are expected to zoom in to at least 150-200% when drawing polygons around complex boundaries (like chair legs or houseplants). Self-Check List: Did I group the left and back walls into a single polygon? (Correction: Split them at the corner). Did I segment through the sofa/table? (Correction: Exclude the sofa/table from the floor/wall polygon). Is there a gap between where the floor ends and the wall begins? (Correction: Ensure the wall and floor boundaries touch seamlessly without overlapping or leaving empty gaps). 5. Deliverables & Format Format: Polygons / Instance Masks (compatible with COCO JSON, Supervise.ly, or CSV polygon coordinates). Sample Output Schema: id: Unique identifier for each polygon. class: wall or floor. instance_id: Unique tracking ID per room element.
Intermediate
Estimated Total Project Earnings
$30
Pay per Label
$0.03/label
Time Requirement
Flexible
Duration
< 1 month
Labelers Needed
4
Description of dataset
Images of indoor rooms
Software
CVATHiring Type
Agency or Individual
Required Location
Global - Any Location
Workload / Schedule
35 hours weekly for 10 days
Software
CVATData Type
Image
Task Types
Segmentation
Subject Matter / Industry
image segmentation
Language
Hindi
Activity on this project
Proposals: 43
Invites sent: 0
Unanswered invites: 0
Share this project
Share link