Nahre
Vanilla computer vision research and prototype package. Lets you dive straight into problem solving mindset. You don't have to worry about tedious stuff.
- loading data
- batch configuration
- record processing
Getting started
pip install nahreHow to use
Data
Put your image files in a single folder. Both flat and nested structures are supported.
Processor
Processor class must implement base Processor interface, which ships with this package.
When batch is executed, every record from data set is ran against process method.
This method must return results which matches next processor's interface on the list.
Example
from pathlib import Path
import numpy as np
from austen import Logger
from degas import FluentImage
from lazy import lazy
from pytest import fixture
from skimage import color, exposure, feature
from nahre import Batch, Processor, execute
from nahre.io import Data
class EdgeProcessor(Processor):
def __init__(self, logger: Logger):
super().__init__(logger)
@lazy
def _description(self):
return 'Any description is better than none.'
def process(self, src: np.ndarray):
with FluentImage(src, self.logger, 'preprocessing') as preprocessed:
preprocessed >> (
color.rgb2gray
) >> (
exposure.rescale_intensity
) >> (
exposure.equalize_adapthist
) >> (
feature.canny
)
return {
'preprocessed': preprocessed.image
}
batch = Batch(
data=Data(Path('data')),
processors=[EdgeProcessor],
log_root=Path('log')
)
execute([batch])
nahre will dump any intermediate images using austen package. Additionally, final results will be dumped in separate folder.
Tests
cd [project-path]
python -m pytest
