Skip to content

Object detection

Readers and writers for object detection have the -od suffix.

Download the 17 flowers object detection VOC XML dataset and extract it.

Plugins#

VOC XML to MS COCO#

The following converts the VOC XML dataset into MS COCO format, a format used by frameworks like Detectron2 or MMDetection:

idc-convert \
  -l INFO \
  from-voc-od \
    -l INFO \
    -i "./voc/*.xml" \
  to-coco-od \
    -l INFO \
    -o ./coco

VOC XML to YOLO (train/val/test splits)#

You can also split the data, e.g., into train, validation and test subsets. The following converts the VOC XML now into YOLO format:

idc-convert \
  -l INFO \
  from-voc-od \
    -l INFO \
    -i "./voc/*.xml" \
  to-yolo-od \
    -l INFO \
    -o ./yolo-split \
    --labels ./yolo-split/labels.txt \
    --split_names train val test \
    --split_ratios 70 15 15

NB: The subsets will be placed into sub-directories according to the split name.