Very minimalistic wrapper for EnlightenGAN inference.
It uses carefully converted pretrained weights (+ baked in preprocessing) from the original repo and only requires onnxruntime as inference engine.
pip install git+https://github.com/arsenyinfo/EnlightenGAN-inference
from enlighten_inference import EnlightenOnnxModel
import cv2
img = cv2.imread('/path/to/image.jpg')
# by default, CUDAExecutionProvider is used
model = EnlightenOnnxModel()
# however, one can choose the providers priority, e.g.:
model = EnlightenOnnxModel(providers = ["CPUExecutionProvider"])
processed = model.predict(img)