RAD Road Anomaly and Actor Detection (Object Detection)

Goal
- Detect and localize road-related objects in images. Classes: HMV, LMV, Pedestrian, RoadDamages, SpeedBump, UnsurfacedRoad.

Public data layout (all paths are under the public/ directory)
- public/train/images/                # Training images
- public/test/images/                 # Test images
- public/train.csv                    # Training annotations (bounding boxes)
- public/test.csv                     # List of test image_ids (exactly one row per image)
- public/sample_submission.csv        # Submission template with required columns
- public/description.txt              # This description file

CSV specifications
- train.csv columns:
    image_id, width, height, class, x_min, y_min, x_max, y_max
  Notes:
    • image_id is the file name of the image under public/train/images/
    • class is one of the six classes listed above
    • Box coordinates are integer pixel indices (inclusive) and within image bounds

- test.csv columns:
    image_id
  Notes:
    • Exactly one row per test image in public/test/images/

- sample_submission.csv columns:
    image_id, PredictionString
  PredictionString encoding:
    Sequence of 6-token groups per detection: "class score x_min y_min x_max y_max"
    Example: "HMV 0.92 10 10 40 40 Pedestrian 0.55 50 60 90 120"
  Notes:
    • The class must be one of: HMV, LMV, Pedestrian, RoadDamages, SpeedBump, UnsurfacedRoad
    • score is a confidence between 0 and 1
    • Coordinates are integer pixel indices (inclusive). If there are no predictions for an image, leave the field empty.

Evaluation summary
- COCO-style mean Average Precision (mAP) averaged over IoU thresholds 0.50:0.05:0.95 and over all classes.
- Submissions must include exactly one row per image listed in public/test.csv.

Getting started
- Train using public/train/images/ and public/train.csv.
- Create predictions for public/test/images/ and save them in the sample_submission.csv format.
- Ensure all image_id values in public/test.csv are present exactly once in your submission.
