Reading the details in a plan.
Applying object detection to large electrical house plans without losing the small symbols.
YOLOv8
Detection model
Patch processing
Handling large drawings
Per-class counts
Turning detections into structured results
A large-image problem
Electrical house plans can be very large images, while the symbols you want to identify are small. Resizing the whole plan down to a model’s input size can make those details difficult to preserve.
I worked on a YOLOv8 detection workflow for electrical house plans, including handling plans around 9600 by 7200 pixels through patches.
What I worked on
The workflow divides a large plan into smaller image patches, runs detection, and brings the results back into the context of the original image. I also worked with per-class counts and merging patch results.
This puts the processing around the model in focus. The coordinates, boundaries, and combined results need to make sense if the output is going to be useful.
Why I find it interesting
Computer vision gives me another way to turn information that is difficult to search into something structured. I like the combination of model work, image processing, and application logic.
It is also a useful reminder that choosing a model is only one part of building a working system. The input pipeline and the way results are presented deserve just as much attention.