Skip to content

Latest commit

 

History

History
36 lines (25 loc) · 1.79 KB

MoFF.md

File metadata and controls

36 lines (25 loc) · 1.79 KB
Raw

Motifs on the Fractured Fresco Fragments (MOFF) dataset

The Motifs on the Fresco Fragments (MOFF) dataset is used as a benchmark for semantic segmentation, and classification tasks for fractured fresco fragments. It contains pixel-wise annotations of 12 motif categories shown below, which were created in Segments.ai.

motif_categories2

Semantic classes

The semantic classes are also defined in the YAML files (in the yolo_processing folder, repair_motif_boxes.yaml for bounding box annotation and repair_motif_seg.yaml for polygonal shapes annotations). In the final mask images the class numbers correspond to the following categories:

  • 0: images_background (image background)
  • 1: fragment_background (the unadorned fragment region)
  • 2: blue_bird,
  • 3: yellow_bird,
  • 4: red_griffon,
  • 5: red_flower,
  • 6: blue_flower,
  • 7: red_circle,
  • 8: red_spiral,
  • 9: curved_green_stripe
  • 10: thin_red_stripe,
  • 11: thick_red_stripe,
  • 12: thin_floral_stripe,
  • 13: thick_floral_stripe

Dataset is published with two schemes of pixel annotations, i.e., for 3-class and 13-class semantic segmentation presented as Scenario 1 and 2 in the paper, respectively.

gts (1)

Illustration of pixel-wise semantic annotations for the MoFF dataset. The input image is on the left, followed by 3-class annotations for Scenario 1 in the middle, and motif-wise annotations for Scenario 2 on the right (different colors represent distinct pixel classes)