instance_segmentation

Instance Segmentation Example

Annotation

labelme data_annotated --labels labels.txt --validatelabel exact --config '{shift_auto_shape_color: -2}'
labelme data_annotated --labels labels.txt --labelflags '{.*: [occluded, truncated], person: [male]}'

Convert to VOC-format Dataset

# It generates:
#   - data_dataset_voc/JPEGImages
#   - data_dataset_voc/SegmentationClass
#   - data_dataset_voc/SegmentationClassNpy
#   - data_dataset_voc/SegmentationClassVisualization
#   - data_dataset_voc/SegmentationObject
#   - data_dataset_voc/SegmentationObjectNpy
#   - data_dataset_voc/SegmentationObjectVisualization
./labelme2voc.py data_annotated data_dataset_voc --labels labels.txt

Fig 1. JPEG image (left), JPEG class label visualization (center), JPEG instance label visualization (right)

Note that the label file contains only very low label values (ex. 0, 4, 14), and 255 indicates the __ignore__ label value (-1 in the npy file).
You can see the label PNG file by following.

../tutorial/draw_label_png.py data_dataset_voc/SegmentationClass/2011_000003.png   # left
../tutorial/draw_label_png.py data_dataset_voc/SegmentationObject/2011_000003.png  # right

Convert to COCO-format Dataset

# It generates:
#   - data_dataset_coco/JPEGImages
#   - data_dataset_coco/annotations.json
./labelme2coco.py data_annotated data_dataset_coco --labels labels.txt