Metrcis for Low Light Enhancement

Here we provide the project able to compute metrics for low light enhancement !

Metrics provided are these:

• MAE
• MSE
• PSNR
• SSIM
• LPIPS
• NIQE
• NIMA

To use it, you need to prepared these packages:

numpy~=1.19.5
pillow~=8.2.0
scipy~=1.7.1
lpips~=0.1.3
filetype~=1.0.7
opencv-python~=4.5.3.56
torch~=1.9.0
torchvision~=0.10.0
tqdm~=4.61.2
scikit-image~=0.18.1

We have proposed requirements.txt so you can run pip install -r requirements.txt.

For NIQE:

Out of scipy.misc.imresize deprecated, in NIQE.py we change

img2 = scipy.misc.imresize(img, 0.5, interp='bicubic', mode='F')

to

from PIL import Image
im = Image.fromarray(img)
size = tuple((np.array(im.size) * 0.5).astype(int))
img2 = np.array(im.resize(size, Image.BICUBIC))

According to this, we think that mode='F' should keep the image original.

---

CLASS Metrics(mode, file_path, name="Metrics", use_gpu=True, A_name='_real_A', B_name='_fake_B', data_path='', A_path='', B_path='')

To run it, you should provide two parameters: mode and file_path.

• data_path: the path to your data folder.

• file_path: the folder path you want to set scores. The scores will be stored in .txt file.

• use_gpu: The default value is True when cuda is available, and of course it will be False when cuda isn't available. If you don't want to use gpu, then you can set it False.

• mode: You must set the mode value.

  'mixed' mode means that your first dataset and your second dataset are mixed but all of first dataset are named by ྾྾྾ and the other are named by ྾྾.By using this mode, you should set the parameter data_path.

  'same_dir_parted' mode means that your tow datasets are stored in two folders and the two folders are storted in the same father folder. And then you should set the parameter data_path. To emphasize it, its kid folders should be sorted alphabetically. The former will be the groundtruth dataset and the latter will be the low-light dataset.

  'diff_dir_parted' mode means that your two datasets are stored in two folders and the two folders are storted in the different folders. So you should set the two parameters: A_path, B_path. And of course the former will be the groundtruth dataset and the latter will be the low-light dataset.

• A_name, B_name: Only when mode is set as 'mixed', both are effective. The default value of A_name is '_real_A' and another is '_fake_B'. Those images of your groundtruth dataset are named by ྾྾྾_real_A and those images of your low-light dataset are named by ྾྾྾_fake_B.

• A_path, B_path: Only when mode is set as 'diff_dir_parted', both are effective. The former will be the path to the groundtruth dataset and the latter will be the path to the low-light dataset.

---

Warning: For those no-reference metics, they will only load second datasets!

---

Now we will show the usage:

import Metrics
metrics = Metrics(mode='mixed', data_path='/home/joy/ParaPata/', file_path='/home/joy/ParaPata/Pata/', A_name='_real', B_name='_fake')
MAE = metrics['MAE']
print(MAE)

Specially, you can use metrics['All'] to record all metrics!

To use the project, you should download these models following by the link and put them in ./models/

If you need to reload datasets, you can call the object to reset some parameters. Here is the example:

import Metrics
metrics = Metrics(mode='mixed', data_path='/home/joy/ParaPata/', file_path='/home/joy/ParaPata/Pata/')
MAE = metrics['MAE']
print(MAE)
metrics(mode='same_dir_parted', file_path='/home/joy/ParaPata/Pata/Olah', data_path='/home/joy/ParaPata/yahoo/')
MSE = metrics['MSE']