public-v1

INSTALL.md

@@ -0,0 +1,28 @@
+# Installation
+
+This repository is built in PyTorch 1.12.0 and tested on Ubuntu 16.04 environment (Python3.7, CUDA10.2, cuDNN7.6).
+Follow these intructions
+
+1. Clone our repository
+```
+git clone https://github.com/INVOKERer/LoFormer.git
+cd AdaRevD
+```
+
+2. Make conda environment
+```
+conda create -n pytorch112 python=3.8
+conda activate pytorch112
+```
+
+3. Install dependencies
+```
+conda install pytorch=1.12.0 torchvision cudatoolkit=11.3 -c pytorch
+pip install matplotlib scikit-learn scikit-image opencv-python yacs joblib natsort h5py tqdm ptflops
+pip install seaborn spectral einops kornia timm gdown addict future lmdb numpy pyyaml requests scipy tb-nightly yapf lpips
+```
+
+4. Install basicsr
+```
+python setup.py develop --no_cuda_ext --record install_files.txt
+```

LICENSE.md

@@ -0,0 +1,29 @@
+## ACADEMIC PUBLIC LICENSE
+
+### Permissions
+:heavy_check_mark: Non-Commercial use
+:heavy_check_mark: Modification
+:heavy_check_mark: Distribution
+:heavy_check_mark: Private use
+
+### Limitations
+:x: Commercial Use
+:x: Liability
+:x: Warranty
+
+### Conditions
+:information_source: License and copyright notice
+:information_source: Same License
+
+LoFormer is free for use in noncommercial settings: at academic institutions for teaching and research use, and at non-profit research organizations.
+You can use LoFormer in your research, academic work, non-commercial work, projects and personal work. We only ask you to credit us appropriately. 
+
+You have the right to use the software, to distribute copies, to receive source code, to change the software and distribute your modifications or the modified software.
+If you distribute verbatim or modified copies of this software, they must be distributed under this license.
+This license guarantees that you're safe when using LoFormer in your work, for teaching or research.
+This license guarantees that LoFormer will remain available free of charge for nonprofit use.
+You can modify LoFormer to your purposes, and you can also share your modifications.
+
+If you would like to use LoFormer in commercial settings, contact us so we can discuss options. Send an email to [email protected]
+
+

Motion_Deblurring/LPIPS.py

@@ -0,0 +1,88 @@
+# coding:utf-8
+import lpips
+import cv2
+import glob
+import tqdm
+
+class util_of_lpips():
+    def __init__(self, net='vgg', use_gpu=False):
+        '''
+        Parameters
+        ----------
+        net: str
+            抽取特征的网络，['alex', 'vgg']
+        use_gpu: bool
+            是否使用GPU，默认不使用
+        Returns
+        -------
+        References
+        -------
+        https://github.com/richzhang/PerceptualSimilarity/blob/master/lpips_2imgs.py
+
+        '''
+        ## Initializing the model
+        self.loss_fn = lpips.LPIPS(net=net)
+        self.use_gpu = use_gpu
+        if use_gpu:
+            self.loss_fn.cuda()
+
+    def calc_lpips(self, img_p):
+        '''
+        Parameters
+        ----------
+        img1_path : str
+            图像1的路径.
+        img2_path : str
+            图像2的路径.
+        Returns
+        -------
+        dist01 : torch.Tensor
+            学习的感知图像块相似度(Learned Perceptual Image Patch Similarity, LPIPS).
+
+        References
+        -------
+        https://github.com/richzhang/PerceptualSimilarity/blob/master/lpips_2imgs.py
+
+        '''
+        img1_path, img2_path = img_p
+        # Load images
+        img0 = lpips.im2tensor(lpips.load_image(img1_path))  # RGB image from [-1,1]
+        img1 = lpips.im2tensor(lpips.load_image(img2_path))
+
+        if self.use_gpu:
+            img0 = img0.cuda()
+            img1 = img1.cuda()
+
+        dist01 = self.loss_fn.forward(img0, img1).item()
+        del img0
+        del img1
+        return dist01
+
+
+if __name__ == '__main__':
+    real_way = '/home/ubuntu/106-48t/personal_data/mxt/Datasets/Deblur/GoPro/test/sharp'
+    fake_Way = '/home/ubuntu/106-48t/personal_data/mxt/exp_results/images_deblur/Deblurring_WNAFNet_32_3stage2b_ffc3gelusin_freqloss/GoPro'
+    img_list_A = glob.glob(real_way + '/*')
+    # img_list_B = glob.glob(fake_Way + '/*')
+    img_list_A.sort()
+    # img_list_B.sort()
+    lpips_list = []
+    # ssim_list = []
+    LPIPS_Mec = util_of_lpips(use_gpu=True)
+    for i in tqdm.tqdm(range(len(img_list_A))):
+
+        # print(img_list_A[i].split('/')[-1], img_list_B[i].split('/')[-1])
+        img_list_B_way = fake_Way + '/' + img_list_A[i].split('/')[-1]
+        # img_list_B_way = glob.glob(fake_Way + '/' + img_list_A[i].split('/')[-1].split('.')[0]+'_*.png')[0]
+        # realim = cv2.imread(img_list_A[i])
+        # fakeim = cv2.imread(img_list_B_way)
+        # iou_s = diceCoeff_score(outim, gtim)
+        # print([img_list_A[i], img_list_B_way])
+        lpips_value = LPIPS_Mec.calc_lpips([img_list_A[i], img_list_B_way])
+        lpips_list.append(lpips_value)
+        # ssim_list.append(ssim)
+    lpips_avr = sum(lpips_list)/len(lpips_list)
+
+    print('lpips_avr')
+    # print(psnr_list)
+    print(lpips_avr)