cv-datakit 0.1.1

Tools to merge and remap computer vision datasets

datakit

Python package for YOLO-format dataset operations:

• merge multiple datasets into one
• merge multiple class names into a target class
• remap class IDs
• visualize labeled samples

CLI Usage

1) Merge datasets

datakit merge /path/ds1 /path/ds2 --out /path/out

2) Merge classes

datakit merge-classes /path/dataset --from Backpack Backpacks --to bag

3) Remap classes

datakit remap /path/dataset --names bag person --map 0:0 1:0 2:1

Remap safety behavior:

• validates that all mapped target IDs are within length of given class range
• pre-scans all label files to ensure every class ID has a mapping before writing
• only writes labels and data.yaml after validation succeeds

4) Visualize samples

datakit visualize --images-dir /path/dataset/val/images --labels-dir /path/dataset/val/labels --n 12 --seed 1

Python API

from datakit import merge_datasets, merge_classes, remap_dataset, plot_random_samples

merge_datasets(["/path/ds1", "/path/ds2"], "/path/out")
merge_classes("/path/dataset", ["Backpack", "Backpacks"], "bag")
remap_dataset("/path/dataset", ["bag", "person"], {0: 0, 1: 0, 2: 1})
plot_random_samples("/path/dataset/val/images", "/path/dataset/val/labels", n=12, seed=1)