Converting datasets

Use argus-cv convert to move between supported segmentation formats.

Supported conversions

• Mask dataset -> YOLO segmentation (--to yolo-seg)
• YOLO segmentation -> COCO (--to coco)
• YOLO segmentation -> Roboflow COCO (--to roboflow-coco)
• YOLO segmentation -> Roboflow COCO RLE (--to roboflow-coco-rle)

1. Mask -> YOLO segmentation

argus-cv convert -i /datasets/road_masks -o /datasets/road_yolo --to yolo-seg

Useful options:

• --epsilon-factor, -e: polygon simplification (default 0.005)
• --min-area, -a: minimum contour area in pixels (default 100)

2. YOLO segmentation -> COCO

argus-cv convert -i /datasets/road_yolo -o /datasets/road_coco --to coco

3. YOLO segmentation -> Roboflow COCO

argus-cv convert -i /datasets/road_yolo -o /datasets/road_rf_coco --to roboflow-coco

4. YOLO segmentation -> Roboflow COCO RLE

argus-cv convert -i /datasets/road_yolo -o /datasets/road_rf_coco_rle --to roboflow-coco-rle

Use this for objects with holes/exclusion zones (donut topology), where polygon-ring interpretation can vary across training libraries.

Output behavior

• Output directory must be empty or not exist.
• Relative output paths are resolved under the input dataset root.
• Argus keeps dataset splits when they exist.

Common errors

• "No MaskDataset found": input is not a mask dataset for --to yolo-seg.
• "YOLO dataset is not a segmentation dataset": source labels are detection, not polygons.
• "Output directory already exists and is not empty": choose another path or empty it first.