dataset-viewer 1.0.1

Allow to view a dataset image by image and edit their labels

Dataset Viewer

A small napari wrapper that lets you step through a folder of images and paint / edit segmentation labels for each one. Labels are saved to disk as PNGs, one per image.

Install

It is recommended to install the dataset viewer into a virtual environment. To install, run:

python -m pip install "dataset_viewer[gui]"

The [gui] extra installs PyQt5 as the napari Qt backend. Omit it if you already have a Qt binding available in your environment.

Python version — developed and tested on Python 3.9. requires-python = ">=3.9" allows newer interpreters, but no testing has been done on 3.10+; use at your own risk.

Data layout

The viewer expects a directory with two sub-folders:

<data_path>/
├── images/     # .jpg, .jpeg or .png source images
└── labels/     # .png label masks, created/updated by the viewer

Labels are always written as PNG (other formats compress and corrupt label values) and named after the source image stem — e.g. images/cat_01.jpg → labels/cat_01.png. An image with an all-zero label array is represented by the absence of the label file; the viewer deletes the file when you clear a mask.

Usage

from pathlib import Path
from dataset_viewer import Viewer, Dataset

Viewer(Dataset(Path("path/to/data"))).start()

Dataset() with no argument falls back to ../../data relative to the current working directory, so prefer passing an explicit path.

Key bindings

Key	Action
Left	Previous image (saves current mask)
Right	Next image (saves current mask)
l	Print the active layer name
Escape	Close the viewer

Painting/erasing uses napari's standard labels-layer controls.

License

MIT — see LICENSE.