README.md

Examples

• one_tag_different_dirs: all files have the same tag, but they are located in different sub-directories
• one_tag_same_dir: all files have the same tag (type) and are located in the same directory
• two_tags_same_dir: two different types of files image and label (two tags), located in the same directory and matched by a fragment of their filenames
• inputs_across_dirs: once again, there are two file tags. Each is located in its own directory, and files are matched by their filename
• one_input_per_dir: files that must be grouped together are located in the same subdirectory
• one_input_per_multi_dirs: same as before, but subdirectories are no longer in the root directory
• different_nesting_levels:
• different_variadicities: the same label file is re-used in different input groups
• different_variadicities_multiple_levels: same as previous, with a more complex layout
• multiple_over_multiple_tags: all tags are multiple (i.e. for a given key, each tag is a list of file rather than a single file)
• optional_tags: all tags are optional (i.e. an input group can have some of its tags to None)
• variadic_files: a single label file and multiple image files for a given key

Kaggle examples

Here are some examples of applying grob to Kaggle datasets. If you find more compelling examples, don't hesitate to add them here.

Rider Safety & Compliance

grob "image={subset}/images/new{id:d}.jpg,labels={subset}/images/new{id:d}.txt" . --optional labels --remove-on-missing image

{
  "train_1": {
    "image": "train/images/new1.jpg",
    "labels": "train/labels/new1.txt"
  }
}

VehicleDetection-YOLOv8

grob "image={subset}/images/{name}.jpg, labels={subset}/labels/{name}.txt" .

{
  "test_00dea1edf14f09ab_jpg.rf.3f17c8790a68659d03b1939a59ccda80": {
    "image": "test/images/00dea1edf14f09ab_jpg.rf.3f17c8790a68659d03b1939a59ccda80.jpg",
    "labels": "test/labels/00dea1edf14f09ab_jpg.rf.3f17c8790a68659d03b1939a59ccda80.txt"
  }
}

B200 LEGO Detection Dataset

grob "image={subset}/images/{name:d}.png, labels={subset}/annotations/{name:d}.png" .

{
  "0": {
    "image": "images/0.png",
    "labels": "annotations/0.xml"
  }
}

Aero-engine defect

grob "image=images/{subset}/images/{name}.jpg, labels=labels/{subset}/{name}.txt" .
# Alternatively, group all the different augmentations of a single image together
grob "image=images/{subset}/images/{id:d}*.jpg, labels=labels/{subset}/{id:d}*.txt" . --multiple --optional

Car Parts - Image Classification

grob "image={subset}/{part_name}/*.jpg" . --multiple --key "{part_name} - {subset}"

ASL Alphabet

grob "image=asl_alphabet_train/{part_name}/*.jpg" . --multiple

Fruit Images for Object Detection

grob "image={subset}/{name}.jpg, labels={subset}/{name}.xml" .

Cat Dataset

grob "image=*/{name}.jpg, labels=*/{name}.jpg.cat" .

DeepGlobe Road Extraction Dataset

grob "image={subset}/{name}_sat.jpg, mask={subset}/{name}_mask.png" .