walwe/autolabel

Autolabel is an image labeling tool. Currently images are labeled using ResNet18-152 implemented by pytorch. Autolabel can be used as a cli tool or as a library.

Autolabel

Autolabel is an image labeling tool. Currently images are labeled using ResNet18-152 implemented by pytorch. Autolabel can be used as a cli tool or as a library.

Installation

python setup.py install

Command line usage

See --help for a command overview

Usage: autolabel [OPTIONS] [IMAGES]...

Options:
  --batch-size INTEGER
  --sep TEXT                      Separator
  --top INTEGER
  -o, --output FILENAME           Output file
  -m, --model [resnet18|resnet34|resnet50|resnet101|resnet152]
  --help                          Show this message and exit.

In the simplest form this mean:

autolabel image.jpg

Autolabel supports reading file names from STDIN:

find /myimages -type f -iname '*.jpg' | autolabel

Library usage

from autolabel.image import ImageListDataset
from autolabel.classifier.resnet import Resnet18Classifier
from pathlib import Path

classifier = Resnet18Classifier()
images = [Path('/path/to/image.jpg'), Path('/path/to/another/image.png')]
dataset = ImageListDataset(images)
res = classifier.predict(dataset, top=top)
for p, decoded in res.items():
    print(p, decoded)