SAGAN: Deep Semantic-Aware Generative Adversarial Network for Unsupervised Image Enhancement [Paper]
This paper has been accepted by Knowledge-Based Systems.
- Python 3.7.13
- Torch 1.12.0
- Visdom 0.1.8.9
- Torchvision 0.13.0
- Numpy 1.21.6
- Pillow 9.2.0
- Onnx 1.13.1
- Onnxruntime 1.13.1
- Training dataset: Unpaired images
- Testing dataset: MEF, LIME, NPE, DICM
Download SAGAN model from Inference model
import numpy as np
import onnxruntime
# load a low-light image
input_img = Image.open('test.png').convert('RGB')
input_img = input_img.resize((600, 400))
transform_list = [transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]
transform = transforms.Compose(transform_list)
input_img = transform(input_img)
input_img = torch.unsqueeze(input_img, 0).numpy()
# predict
inference = onnxruntime.InferenceSession('model.onnx')
input_img = {'input': input_img}
output_image = inference.run(['output'], input_img)[0]
# save enhanced image
output_image = np.transpose(output_image[0], (1, 2, 0))
output_image = np.clip(output_image, 0, 255).astype(np.uint8)
output_image = Image.fromarray(output_image)
output_image.save('result.png')