About run.ipynb

**bbox_per_image.yolo_labels(mask_dir='resources/Building/masks/labels_11m_75', save_dir='resources/Building/bbox_perslice/YOLO-v11m-75p')**

What does this code do?

Converts polygons to Oriented Bounding Boxes

• mask_dir ⇒ Where the masks data (labels, polygons) are stored after model inference
• save_dir ⇒ Where you store the OBBs for each of the mask data (label) The new format for the data is: classid x1 y1 x2 y2 x3 y3 x4 y4 conf_score

You can see from the image that, there is a labels.txt file for each of the slice (for building segmentation)

**bbox_combined.process_sliding_window(label_dir='resources/Building/bbox_perslice/YOLO-v11m-75p', save_dir='resources/ClassifierOutput/Building Classification/YOLO-v11m-75p', SEG_MODEL='yolo')**

What does this code do?

Combines the bounding boxes across windows. (f.e. A building may span multiple windows, merge OBBs to represent a building by a single Bounding Box) [NB: Requires HyperParam Tuning for the best result]

• label_dir ⇒ Where the OBBs are stored
• save_dir ⇒ Where you store the merged OBBs (in full OP dimension)

And a sample label format

You can see the values of the labels. It’s in full orthophoto dimension not in slice dimension.

labels.txt ⇒ Just all the buildings of the OP labels_binclass.txt ⇒ Buildings after primary classficiation

**model = 'YOLO-v11m-25p'

binary_classifier.classify(building_labels_path=f'resources/ClassifierOutput/Building Classification/{model}/labels.txt', object_labels_path=f'resources/ClassifierOutput/Object Detection/YOLO-v9c/predicted_object_labels_v9gelanc.txt', output_labels_path=f'resources/ClassifierOutput/Building Classification/{model}/labels_binclass.txt')**

What does this code do?

Primary Classification If there is at least one object within the surface area of an OBB (OR ≥ 80%) then mark that OBB (building) as Risky (class_id = 1)

• building_labels_path ⇒ Remember where you saved the full OP dimensional OBBs?
• object_labels_path ⇒ Dengue breeding factor objects (6 in our case) represented by bounding boxes (axis aligned) in the full orthphoto dimension. Use this code: utils/object_labels_global_converter.py
• output_labels_path ⇒ Where you store the OBBs (buildings) after the primary classification

Notice: Now there is a class_id for each of the OBB

Final distance-based classification

You have to use this code for the final classification

• Expected Output: Performance graphs. If you want, you can also save the building labels after the final classification for visualization on the original OP.

This code, does the classification along with saving the buildings labels after classfication.

✅ That’s it !