data.yaml

# AeroScapes — Ultralytics YOLO semantic segmentation dataset config
#
# This file lives INSIDE the dataset root itself (next to images/ and masks/)
# so the whole folder is self-contained and portable — no absolute machine
# path to edit. There is intentionally NO 'path:' key: when omitted,
# Ultralytics resolves train/val relative to this file's own directory
# (rather than joining a relative 'path:' against the global DATASETS_DIR
# setting), so this works correctly wherever the folder is placed or
# downloaded to.
#
# Format: https://docs.ultralytics.com/datasets/semantic/
# Masks:  single-channel PNG, pixel value = class index (0-11). 255 is
#         reserved for genuinely unrecognized values (none expected here —
#         see the source repo's convert_aeroscapes_to_yolo.py, which
#         validates every mask pixel at conversion time).

train: images/train # 2621 images
val: images/val # 648 images
# No 'test:' key — the source AeroScapes distribution has no test split
# (only ImageSets/trn.txt and val.txt).

masks_dir: masks # mirrors images/ structure: images/train -> masks/train, etc.

nc: 12

names:
  0: Background
  1: Person
  2: Bike
  3: Car
  4: Drone
  5: Boat
  6: Animal
  7: Obstacle
  8: Construction
  9: Vegetation
  10: Road
  11: Sky

# Example training command (run with this file's directory as --dst, or
# point data= at this file's absolute path from anywhere):
#   yolo semantic train model=yolo26n-sem.pt data=data.yaml imgsz=896 epochs=120 batch=8