Merge pull request #1230 from ananyakaligal/main

Added SRGAN

Author: sanjay-kv

Algorithms and Deep Learning Models/SRGAN for upscaling image resolutions/.gitignore

@@ -0,0 +1,104 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# pyenv
+.python-version
+
+# celery beat schedule file
+celerybeat-schedule
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/

Algorithms and Deep Learning Models/SRGAN for upscaling image resolutions/README.md

@@ -0,0 +1,18 @@
+This is basically a Super Resolution Generative Adversarial Network (SRGAN) with the purpose of upscaling image resolutions by a factor of two using deep learning. This way, a picture which initially appears pixellated and/or blurry can be modified so that the features are quite more distinguishable. The model is trained on the COCO unlabeled2017 dataset. Download [here](http://cocodataset.org/#download).
+
+## Requirements
+- Tensorflow 2.0
+- Scipy, Numpy
+- PIL
+- Matplotlib
+- MS COCO unlabeled2017 Dataset (for training)
+
+## Usage
+To train model (which we highly reccomend doing some more):
+```
+python srgan.py
+```
+To run the model on an image:
+```
+python srgan.py -p image.jpg
+```

Algorithms and Deep Learning Models/SRGAN for upscaling image resolutions/main/data_loader.py

@@ -0,0 +1,66 @@
+import scipy.misc
+from glob import glob
+import numpy as np
+import matplotlib.pyplot as plt
+
+class DataLoader():
+    def __init__(self, dataset_name, img_res=(200, 200)):
+        self.dataset_name = dataset_name
+        self.img_res = img_res
+    def load_data(self, batch_size=1, is_testing=False):
+        data_type = "train" if not is_testing else "test"
+
+        path = glob('../../datasets/%s/*' % (self.dataset_name))
+
+        batch_images = np.random.choice(path, size=batch_size)
+
+        imgs_hr = []
+        imgs_lr = []
+        for img_path in batch_images:
+            img = self.imread(img_path)
+
+            h, w = self.img_res
+            low_h, low_w = int(h / 2), int(w / 2)
+
+            img_hr = scipy.misc.imresize(img, self.img_res)
+            img_lr = scipy.misc.imresize(img, (low_h, low_w))
+
+            # If training => do random flip
+            if not is_testing and np.random.random() < 0.5:
+                img_hr = np.fliplr(img_hr)
+                img_lr = np.fliplr(img_lr)
+
+            imgs_hr.append(img_hr)
+            imgs_lr.append(img_lr)
+
+        imgs_hr = np.array(imgs_hr) / 127.5 - 1.
+        imgs_lr = np.array(imgs_lr) / 127.5 - 1.
+        return imgs_hr, imgs_lr
+    def load_pred(self, path):
+        img = self.imread(path)
+        imgs_hr = []
+        imgs_lr = []
+        h, w = self.img_res
+        low_h, low_w = int(h / 2), int(w / 2)
+        img_hr = scipy.misc.imresize(img, (self.img_res))
+        img_lr = scipy.misc.imresize(img, (low_h, low_w))
+        imgs_hr.append(img_hr)
+        imgs_lr.append(img_lr)
+        imgs_hr = np.array(imgs_hr) / 127.5 - 1.
+        imgs_lr = np.array(imgs_lr) / 127.5 - 1.
+        return imgs_hr, imgs_lr
+
+    def load_resize(self, path):
+        img = self.imread(path)
+        imgs_hr = []
+        imgs_lr = []
+        h, w = self.img_res
+        low_h, low_w = int(h/2), int(w/2)
+        img_hr = scipy.misc.imresize(img, (self.img_res))
+        img_lr = scipy.misc.imresize(img, (img_lr))
+        imgs_hr = np.array(imgs_hr) / 127.5 - 1.
+        imgs_lr = np.array(imgs_lr) / 127.5 - 1.
+        imgs_hr = np.resize(imgs_hr, (-1, 400,400,3))
+
+    def imread(self, path):
+        return scipy.misc.imread(path, mode='RGB').astype(np.float)

Algorithms and Deep Learning Models/SRGAN for upscaling image resolutions/main/preprocessing.py

@@ -0,0 +1,146 @@
+import numpy as np
+import matplotlib.pyplot as plt
+import os
+import math
+path = '../../datasets/unlabeled2017/000000002272.jpg'
+
+class ImageSlicer(object):
+    def __init__(self, source, size, strides=[None, None], BATCH = False, PADDING=False):
+        self.source = source
+        self.size = size
+        self.strides = strides
+        self.BATCH = BATCH
+        self.PADDING = PADDING
+
+    def __read_images(self):
+        Images = []
+        image_names = sorted(os.listdir(self.source))
+        for im in image_names:
+            image = plt.imread(os.path.join(dir_path,im))
+            Images.append(image)
+        return Images
+
+    def __offset_op(self, input_length, output_length, stride):
+        offset = (input_length) - (stride*((input_length - output_length)//stride)+output_length)
+        return offset
+
+    def __padding_op(self, Image):
+        if self.offset_x > 0:
+            padding_x = self.strides[0] - self.offset_x
+        else:
+            padding_x = 0
+        if self.offset_y > 0:
+            padding_y = self.strides[1] - self.offset_y
+        else:
+            padding_y = 0
+        Padded_Image = np.zeros(shape=(Image.shape[0]+padding_x, Image.shape[1]+padding_y, Image.shape[2]),dtype=Image.dtype)
+        Padded_Image[padding_x//2:(padding_x//2)+(Image.shape[0]),padding_y//2:(padding_y//2)+Image.shape[1],:] = Image
+        return Padded_Image
+
+    def __convolution_op(self, Image):
+        start_x = 0
+        start_y = 0
+        self.n_rows = Image.shape[0]//self.strides[0] + 1
+        self.n_columns = Image.shape[1]//self.strides[1] + 1
+        # print(str(self.n_rows)+" rows")
+        # print(str(self.n_columns)+" columns")
+        small_images = []
+        for i in range(self.n_rows-1):
+            for j in range(self.n_columns-1):
+                new_start_x = start_x+i*self.strides[0]
+                new_start_y= start_y+j*self.strides[1]
+                small_images.append(Image[new_start_x:new_start_x+self.size[0],new_start_y:new_start_y+self.size[1],:])
+        return small_images
+
+    def transform(self):
+
+        if not(os.path.exists(self.source)):
+            raise Exception("Path does not exist!")
+
+        else:
+            if self.source and not(self.BATCH):
+                Image = plt.imread(self.source)
+                Images = [Image]
+            else:
+                Images = self.__read_images()
+
+            im_size = Images[0].shape
+            num_images = len(Images)
+            transformed_images = dict()
+            Images = np.array(Images)
+
+            if self.PADDING:
+
+                padded_images = []
+
+                if self.strides[0]==None and self.strides[1]==None:
+                    self.strides[0] = self.size[0]
+                    self.strides[1] = self.size[1]
+                    self.offset_x = Images.shape[1]%self.size[0]
+                    self.offset_y = Images.shape[2]%self.size[1]
+                    padded_images = list(map(self.__padding_op, Images))
+
+                elif self.strides[0]==None and self.strides[1]!=None:
+                    self.strides[0] = self.size[0]
+                    self.offset_x = Images.shape[1]%self.size[0]
+                    if self.strides[1] <= Images.shape[2]:
+                        self.offset_y = self.__offset_op(Images.shape[2], self.size[1], self.strides[1])
+                    else:
+                        raise Exception("stride_y must be between {0} and {1}".format(1,Images.shape[2]))
+                    padded_images = list(map(self.__padding_op, Images))
+
+                elif self.strides[0]!=None and self.strides[1]==None:
+                    self.strides[1] = self.size[1]
+                    self.offset_y = Images.shape[2]%self.size[1]
+                    if self.strides[0] <=Images.shape[1]:
+                        self.offset_x = self.__offset_op(Images.shape[1], self.size[0], self.strides[0])
+                    else:
+                        raise Exception("stride_x must be between {0} and {1}".format(1,Images.shape[1]))
+                    padded_images = list(map(self.__padding_op, Images))
+
+                else:
+                    if self.strides[0] > Images.shape[1]:
+                        raise Exception("stride_x must be between {0} and {1}".format(1,Images.shape[1]))
+
+                    elif self.strides[1] > Images.shape[2]:
+                        raise Exception("stride_y must be between {0} and {1}".format(1,Images.shape[2]))
+
+                    else:
+                        self.offset_x = self.__offset_op(Images.shape[1], self.size[0], self.strides[0])
+                        self.offset_y = self.__offset_op(Images.shape[2], self.size[1], self.strides[1])
+                        padded_images = list(map(self.__padding_op, Images))
+
+                for i, Image in enumerate(padded_images):
+                    transformed_images[str(i)] = self.__convolution_op(Image)
+
+            else:
+                if self.strides[0]==None and self.strides[1]==None:
+                    self.strides[0] = self.size[0]
+                    self.strides[1] = self.size[1]
+
+                elif self.strides[0]==None and self.strides[1]!=None:
+                    if self.strides[1] > Images.shape[2]:
+                        raise Exception("stride_y must be between {0} and {1}".format(1,Images.shape[2]))
+                    self.strides[0] = self.size[0]
+
+                elif self.strides[0]!=None and self.strides[1]==None:
+                    if self.strides[0] > Images.shape[1]:
+                        raise Exception("stride_x must be between {0} and {1}".format(1,Images.shape[1]))
+                    self.strides[1] = self.size[1]
+                else:
+                    if self.strides[0] > Images.shape[1]:
+                        raise Exception("stride_x must be between {0} and {1}".format(1,Images.shape[1]))
+                    elif self.strides[1] > Images.shape[2]:
+                        raise Exception("stride_y must be between {0} and {1}".format(1,Images.shape[2]))
+
+                for i, Image in enumerate(Images):
+                    transformed_images[str(i)] = self.__convolution_op(Image)
+
+            return transformed_images
+
+    def save_images(self,transformed):
+        self.r,self.c = self.n_rows-1, self.n_columns-1
+        for key, val in transformed.items():
+            val = np.array(val, dtype=np.float64) /127.5 -1.
+            val = .5 * val + 0.5
+        return val