Sky Segmentation & Mask Preparation for Sky Replacement

Sample frames for this project are available under NDA. Write to bilal@wortel.ai and we will share the relevant before and after pairs.

Challenges

Handling complex boundaries around trees, buildings and utility poles.
Fine detail had to be preserved while segmentation artefacts were removed.
Producing clean masks for images captured under different lighting and weather conditions.
Maintaining consistent mask quality across the entire dataset.

Project overview

A sky segmentation dataset was prepared for an AI-powered sky replacement pipeline. The client supplied property images captured under poor weather and lighting, where a dull or unattractive sky reduced the overall visual appeal of the photograph.

The objective was to isolate the sky region accurately by generating refined binary masks, so that a realistic sky, such as a blue sky with white clouds, could be inserted while the rest of the scene was preserved. Property and outdoor images were taken as input and refined binary sky masks with alpha channel extraction were produced as output.

Dataset

Images: Not stated
Classes: Not stated
Annotation: Binary mask segmentation
Source: Client-provided
Tool: Python and computer vision models
Client: Private client

Industry

Real Estate & Interiors Media & Privacy

Annotation type

Segmentation / Polygons

Labels

Sky

Related work

Face Blur & Picture Frame Detection

Picture frames masked and faces boxed in property photos, separating real faces from artwork for automatic privacy blurring.

Licence Plate Detection

Rotated four-point polygons around licence plates as small as 12 px, for automatic blurring in property photography.

Bathroom Surface Segmentation

Tiling, sanitaryware and vanity units separated per surface, through the glare and reflections bathrooms are full of.