hi i am working on building Plans 6000 images. i want to detect walls on these plans. i used yolov5 already but mAP50 was about 40% , That’s why i use Yolov11l Now. i set 60 epochs and mAP50 is 69% .
of course i reduced images to 2192 images just now to accelerate training. mAP50 is about 69% for 2192 images. i downloaded building plans dataset from Roboflow website (6000 images and related text label files)
How to improve mAP50 to about 100%?
Depending upon the complexity of your task, it may never reach mAP50 close to 100%. Especially if your dataset and labels are not good quality, which is usually the case with public datasets on Roboflow.
You should clean your dataset and fix any labelling issues. Remove any offline augmentations performed by Roboflow.
How clean my dataset and fix any label issue?
You can go through them and see if there are any labelling issues. The labels should be consistent.
You can read this: