UMSN Files

datasets/__init__.py

@@ -0,0 +1 @@
+

datasets/classification.py

@@ -0,0 +1,75 @@
+import torch.utils.data as data
+from PIL import Image
+import os
+import os.path
+import numpy as np
+import h5py
+import glob
+
+IMG_EXTENSIONS = [
+  '.jpg', '.JPG', '.jpeg', '.JPEG',
+  '.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP',
+]
+
+def is_image_file(filename):
+  return any(filename.endswith(extension) for extension in IMG_EXTENSIONS)
+
+def make_dataset(dir):
+  images = []
+  if not os.path.isdir(dir):
+    raise Exception('Check dataroot')
+  for root, _, fnames in sorted(os.walk(dir)):
+    for fname in fnames:
+      if is_image_file(fname):
+        path = os.path.join(dir, fname)
+        item = path
+        images.append(item)
+  return images
+
+def default_loader(path):
+  return Image.open(path).convert('RGB')
+
+class classification(data.Dataset):
+  def __init__(self, root, transform=None, loader=default_loader, seed=None):
+    # imgs = make_dataset(root)
+    # if len(imgs) == 0:
+    #   raise(RuntimeError("Found 0 images in subfolders of: " + root + "\n"
+    #              "Supported image extensions are: " + ",".join(IMG_EXTENSIONS)))
+    self.root = root
+    # self.imgs = imgs
+    self.transform = transform
+    self.loader = loader
+
+    if seed is not None:
+      np.random.seed(seed)
+
+  def __getitem__(self, _):
+    index = np.random.randint(1,self.__len__())
+    # path = self.imgs[index]
+    # img = self.loader(path)
+    #img = img.resize((w, h), Image.BILINEAR)
+
+
+
+    file_name=self.root+'/'+str(index)+'.h5'
+    f=h5py.File(file_name,'r')
+
+    haze_image=f['haze'][:]
+    label=f['label'][:]
+    label=label.mean()-1
+
+    haze_image=np.swapaxes(haze_image,0,2)
+    haze_image=np.swapaxes(haze_image,1,2)
+
+
+    # if self.transform is not None:
+    #   # NOTE preprocessing for each pair of images
+    #   imgA, imgB = self.transform(imgA, imgB)
+    return haze_image, label
+
+  def __len__(self):
+    train_list=glob.glob(self.root+'/*h5')
+    # print len(train_list)
+    return len(train_list)
+
+    # return len(self.imgs)

datasets/pix2pix.py

@@ -0,0 +1,133 @@
+import torch.utils.data as data
+
+from PIL import Image
+import os
+import os.path
+import numpy as np
+
+
+
+IMG_EXTENSIONS = [
+  '.jpg', '.JPG', '.jpeg', '.JPEG',
+  '.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP', '',
+]
+
+def is_image_file(filename):
+  return any(filename.endswith(extension) for extension in IMG_EXTENSIONS)
+
+def make_dataset(dir):
+  images = []
+  if not os.path.isdir(dir):
+    raise Exception('Check dataroot')
+  for root, _, fnames in sorted(os.walk(dir)):
+    for fname in fnames:
+      if is_image_file(fname):
+        path = os.path.join(dir, fname)
+        item = path
+        images.append(item)
+  return images
+
+def default_loader(path):
+  return Image.open(path).convert('RGB')
+
+class pix2pix(data.Dataset):
+  def __init__(self, root, transform=None, loader=default_loader, seed=None):
+    imgs = make_dataset(root)
+    if len(imgs) == 0:
+      raise(RuntimeError("Found 0 images in subfolders of: " + root + "\n"
+                 "Supported image extensions are: " + ",".join(IMG_EXTENSIONS)))
+    self.root = root
+    self.imgs = imgs
+    self.transform = transform
+    self.loader = loader
+
+    if seed is not None:
+      np.random.seed(seed)
+
+  def __getitem__(self, index):
+    # index = np.random.randint(self.__len__(), size=1)[0]
+    # index = np.random.randint(self.__len__(), size=1)[0]+1
+    # index = np.random.randint(self.__len__(), size=1)[0]
+
+    # index_folder = np.random.randint(1,4)
+    index_folder = np.random.randint(0,1)
+
+    index_sub = np.random.randint(2, 5)
+
+    label=index_folder
+
+
+    if index_folder==0:
+        path='/home/openset/Desktop/derain2018/facades/training2'+'/'+str(index)+'.jpg'
+
+
+
+    if index_folder==1:
+      if index_sub<4:
+        path='/home/openset/Desktop/derain2018/facades/DB_Rain_new/Rain_Heavy/train2018new'+'/'+str(index)+'.jpg'
+      if index_sub==4:
+        index = np.random.randint(0,400)
+        path='/home/openset/Desktop/derain2018/facades/DB_Rain/Rain_Heavy/trainnew'+'/'+str(index)+'.jpg'
+
+    if index_folder==2:
+      if index_sub<4:
+        path='/home/openset/Desktop/derain2018/facades/DB_Rain_new/Rain_Medium/train2018new'+'/'+str(index)+'.jpg'
+      if index_sub==4:
+        index = np.random.randint(0,400)
+        path='/home/openset/Desktop/derain2018/facades/DB_Rain/Rain_Medium/trainnew'+'/'+str(index)+'.jpg'
+
+    if index_folder==3:
+      if index_sub<4:
+        path='/home/openset/Desktop/derain2018/facades/DB_Rain_new/Rain_Light/train2018new'+'/'+str(index)+'.jpg'
+      if index_sub==4:
+        index = np.random.randint(0,400)
+        path='/home/openset/Desktop/derain2018/facades/DB_Rain/Rain_Light/trainnew'+'/'+str(index)+'.jpg'
+
+
+
+    # img = self.loader(path)
+
+    img = self.loader(path)
+
+    # NOTE: img -> PIL Image
+    # w, h = img.size
+    # w, h = 1024, 512
+    # img = img.resize((w, h), Image.BILINEAR)
+    # pix = np.array(I)
+    #
+    # r = 16
+    # eps = 1
+    #
+    # I = img.crop((0, 0, w/2, h))
+    # pix = np.array(I)
+    # base=guidedfilter(pix, pix, r, eps)
+    # base = PIL.Image.fromarray(numpy.uint8(base))
+    #
+    #
+    #
+    # imgA=base
+    # imgB=I-base
+    # imgC = img.crop((w/2, 0, w, h))
+
+    w, h = img.size
+    # img = img.resize((w, h), Image.BILINEAR)
+
+
+    # NOTE: split a sample into imgA and imgB
+    imgA = img.crop((0, 0, w/2, h))
+    # imgC = img.crop((2*w/3, 0, w, h))
+
+    imgB = img.crop((w/2, 0, w, h))
+
+
+    if self.transform is not None:
+      # NOTE preprocessing for each pair of images
+      # imgA, imgB, imgC = self.transform(imgA, imgB, imgC)
+      imgA, imgB = self.transform(imgA, imgB)
+
+    return imgA, imgB, label
+
+  def __len__(self):
+    # return 679
+    print(len(self.imgs))
+    return len(self.imgs)

datasets/pix2pix2.py

@@ -0,0 +1,109 @@
+import torch.utils.data as data
+from PIL import Image
+import os
+import os.path
+import numpy as np
+import h5py
+import glob
+import scipy.ndimage
+IMG_EXTENSIONS = [
+  '.jpg', '.JPG', '.jpeg', '.JPEG',
+  '.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP',
+]
+
+def is_image_file(filename):
+  return any(filename.endswith(extension) for extension in IMG_EXTENSIONS)
+
+def make_dataset(dir):
+  images = []
+  if not os.path.isdir(dir):
+    raise Exception('Check dataroot')
+  for root, _, fnames in sorted(os.walk(dir)):
+    for fname in fnames:
+      if is_image_file(fname):
+        path = os.path.join(dir, fname)
+        item = path
+        images.append(item)
+  return images
+
+def default_loader(path):
+  return Image.open(path).convert('RGB')
+
+class pix2pix(data.Dataset):
+  def __init__(self, root, transform=None, loader=default_loader, seed=None):
+    # imgs = make_dataset(root)
+    # if len(imgs) == 0:
+    #   raise(RuntimeError("Found 0 images in subfolders of: " + root + "\n"
+    #              "Supported image extensions are: " + ",".join(IMG_EXTENSIONS)))
+    self.root = root
+    # self.imgs = imgs
+    self.transform = transform
+    self.loader = loader
+
+    if seed is not None:
+      np.random.seed(seed)
+
+  def __getitem__(self, _):
+    index = np.random.randint(1,self.__len__())
+    # index = np.random.randint(self.__len__(), size=1)[0]
+
+    # path = self.imgs[index]
+    # img = self.loader(path)
+    #img = img.resize((w, h), Image.BILINEAR)
+
+
+
+    file_name=self.root+'/'+str(index)+'.h5'
+    f=h5py.File(file_name,'r')
+
+    haze_image=f['haze'][:]
+    trans_map=f['trans'][:]
+    ato_map=f['ato'][:]
+    GT=f['gt'][:]
+
+
+
+    haze_image=np.swapaxes(haze_image,0,2)
+    trans_map=np.swapaxes(trans_map,0,2)
+    ato_map=np.swapaxes(ato_map,0,2)
+    GT=np.swapaxes(GT,0,2)
+
+
+
+    haze_image=np.swapaxes(haze_image,1,2)
+    trans_map=np.swapaxes(trans_map,1,2)
+    ato_map=np.swapaxes(ato_map,1,2)
+    GT=np.swapaxes(GT,1,2)
+
+    # if np.random.uniform()>0.5:
+    #   haze_image=np.flip(haze_image,2).copy()
+    #   GT = np.flip(GT, 2).copy()
+    #   trans_map=np.flip(trans_map, 2).copy()
+    # if np.random.uniform()>0.5:
+    #   angle = np.random.uniform(-10, 10)
+    #   haze_image=scipy.ndimage.interpolation.rotate(haze_image, angle)
+    #   GT = scipy.ndimage.interpolation.rotate(GT, angle)
+
+    # if np.random.uniform()>0.5:
+    #   angle = np.random.uniform(-10, 10)
+    #   haze_image=scipy.ndimage.interpolation.rotate(haze_image, angle)
+    #   GT = scipy.ndimage.interpolation.rotate(GT, angle)
+
+    # if np.random.uniform()>0.5:
+    #   std = np.random.uniform(0.2, 1.2)
+    #   haze_image = scipy.ndimage.filters.gaussian_filter(haze_image, std,mode='constant')
+
+    #   haze_image=np.random.uniform(-10/5000,10/5000,size=haze_image.shape)
+    #   haze_image = np.maximum(0, haze_image)
+
+    # if self.transform is not None:
+    #   # NOTE preprocessing for each pair of images
+    #   imgA, imgB = self.transform(imgA, imgB)
+    return haze_image, GT,  trans_map, ato_map
+
+  def __len__(self):
+    train_list=glob.glob(self.root+'/*h5')
+    # print len(train_list)
+    return len(train_list)
+
+    # return len(self.imgs)

datasets/pix2pix_class.py

@@ -0,0 +1,88 @@
+import torch.utils.data as data
+
+from PIL import Image
+import os
+import os.path
+import numpy as np
+
+
+IMG_EXTENSIONS = [
+  '.jpg', '.JPG', '.jpeg', '.JPEG',
+  '.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP', '',
+]
+
+def is_image_file(filename):
+  return any(filename.endswith(extension) for extension in IMG_EXTENSIONS)
+
+def make_dataset(dir):
+  images = []
+  if not os.path.isdir(dir):
+    raise Exception('Check dataroot')
+  for root, _, fnames in sorted(os.walk(dir)):
+    for fname in fnames:
+      if is_image_file(fname):
+        path = os.path.join(dir, fname)
+        item = path
+        images.append(item)
+  return images
+
+def default_loader(path):
+  return Image.open(path).convert('RGB')
+
+class pix2pix(data.Dataset):
+  def __init__(self, root, transform=None, loader=default_loader, seed=None):
+    imgs = make_dataset(root)
+    if len(imgs) == 0:
+      raise(RuntimeError("Found 0 images in subfolders of: " + root + "\n"
+                 "Supported image extensions are: " + ",".join(IMG_EXTENSIONS)))
+    self.root = root
+    self.imgs = imgs
+    self.transform = transform
+    self.loader = loader
+
+    if seed is not None:
+      np.random.seed(seed)
+
+  def __getitem__(self, index):
+
+    index_sub = np.random.randint(0, 3)
+    label=index_sub
+
+
+
+    if index_sub==0:
+        index = np.random.randint(0,4000)
+        path='/media/openset/Z/derain2018/facades/DID-MDN-training/Rain_Heavy/train2018new'+'/'+str(index)+'.jpg'
+
+
+    if index_sub==1:
+        index = np.random.randint(0,4000)
+        path='/media/openset/Z/derain2018/facades/DID-MDN-training/Rain_Medium/train2018new'+'/'+str(index)+'.jpg'
+
+    if index_sub==2:
+        index = np.random.randint(0,4000)
+        path='/media/openset/Z/derain2018/facades/DID-MDN-training/Rain_Light/train2018new'+'/'+str(index)+'.jpg'
+
+
+
+    img = self.loader(path)
+
+
+    w, h = img.size
+
+
+    # NOTE: split a sample into imgA and imgB
+    imgA = img.crop((0, 0, w/2, h))
+    imgB = img.crop((w/2, 0, w, h))
+
+
+    if self.transform is not None:
+      # NOTE preprocessing for each pair of images
+      imgA, imgB = self.transform(imgA, imgB)
+
+    return imgA, imgB, label
+
+  def __len__(self):
+    # return 679
+    # print(len(self.imgs))
+    return len(self.imgs)

datasets/pix2pix_val.py

@@ -0,0 +1,139 @@
+import torch.utils.data as data
+
+from PIL import Image
+import os
+import os.path
+import numpy as np
+import cv2
+
+# from guidedfilter import guidedfilter
+# import guidedfilter.guidedfilter as guidedfilter
+
+
+
+
+IMG_EXTENSIONS = [
+  '.jpg', '.JPG', '.jpeg', '.JPEG',
+  '.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP', '',
+]
+
+def is_image_file(filename):
+  return any(filename.endswith(extension) for extension in IMG_EXTENSIONS)
+
+def make_dataset(dir):
+  images = []
+  if not os.path.isdir(dir):
+    raise Exception('Check dataroot')
+  for root, _, fnames in sorted(os.walk(dir)):
+    for fname in fnames:
+      if is_image_file(fname):
+        path = os.path.join(dir, fname)
+        item = path
+        images.append(item)
+  return images
+
+def default_loader(path):
+  return Image.open(path).convert('RGB')
+
+class pix2pix_val(data.Dataset):
+  def __init__(self, root, transform=None, loader=default_loader, seed=None):
+    imgs = make_dataset(root)
+    if len(imgs) == 0:
+      raise(RuntimeError("Found 0 images in subfolders of: " + root + "\n"
+                 "Supported image extensions are: " + ",".join(IMG_EXTENSIONS)))
+    self.root = root
+    self.imgs = imgs
+    self.transform = transform
+    self.loader = loader
+
+    if seed is not None:
+      np.random.seed(seed)
+
+  def __getitem__(self, index):
+    # index = np.random.randint(self.__len__(), size=1)[0]
+    # index = np.random.randint(self.__len__(), size=1)[0]
+
+    path = self.imgs[index]
+    
+    if index>75000:
+        index = index%75000+10
+    path=self.root+'/'+'%06d'%(index+1)+'.png'
+
+    index_folder = np.random.randint(0,4)
+    label=index_folder
+
+    # path='/home/openset/Desktop/derain2018/facades/DB_Rain_test/Rain_Heavy/test2018'+'/'+str(index)+'.jpg'
+    img = self.loader(path)
+
+    # # NOTE: img -> PIL Image
+    # w, h = img.size
+    # w, h = 1024, 512
+    # img = img.resize((w, h), Image.BILINEAR)
+    # # NOTE: split a sample into imgA and imgB
+    # imgA = img.crop((0, 0, w/2, h))
+    # imgB = img.crop((w/2, 0, w, h))
+    # if self.transform is not None:
+    #   # NOTE preprocessing for each pair of images
+    #   imgA, imgB = self.transform(imgA, imgB)
+    # return imgA, imgB
+
+
+    # w, h = 1536, 512
+    # img = img.resize((w, h), Image.BILINEAR)
+    #
+    #
+    # # NOTE: split a sample into imgA and imgB
+    # imgA = img.crop((0, 0, w/3, h))
+    # imgC = img.crop((2*w/3, 0, w, h))
+    #
+    # imgB = img.crop((w/3, 0, 2*w/3, h))
+
+    # w, h = 1024, 512
+    # img = img.resize((w, h), Image.BILINEAR)
+    #
+    # r = 16
+    # eps = 1
+    #
+    # # I = img.crop((0, 0, w/2, h))
+    # # pix = np.array(I)
+    # # print
+    # # base[idx,:,:,:]=guidedfilter(pix[], pix[], r, eps)
+    # # base[]=guidedfilter(pix[], pix[], r, eps)
+    # # base[]=guidedfilter(pix[], pix[], r, eps)
+    #
+    #
+    # # base = PIL.Image.fromarray(numpy.uint8(base))
+    #
+    # # NOTE: split a sample into imgA and imgB
+    # imgA = img.crop((0, 0, w/3, h))
+    # imgC = img.crop((2*w/3, 0, w, h))
+    #
+    # imgB = img.crop((w/3, 0, 2*w/3, h))
+    # imgA=base
+    # imgB=I-base
+    # imgC = img.crop((w/2, 0, w, h))
+    w, h = img.size
+    #print(w,h)
+    # w, h = 586*2, 586
+
+    # img = img.resize((w, h), Image.BILINEAR)
+
+
+    # NOTE: split a sample into imgA and imgB
+    imgA = img.crop((0, 0, w/2, h))
+    # imgC = img.crop((2*w/3, 0, w, h))
+
+    imgB = img.crop((w/2, 0, w, h))
+
+
+
+    if self.transform is not None:
+      # NOTE preprocessing for each pair of images
+      # imgA, imgB, imgC = self.transform(imgA, imgB, imgC)
+      imgA, imgB = self.transform(img, img)
+      
+
+    return imgA, imgB
+
+  def __len__(self):
+    return len(self.imgs)