Merge pull request #6 from bachsh/master

Thank you very much to the two of you !
Yes, opencv-python is sometimes tricky to install, however works most of the times, I will therefore merge this first and then add it to the requirements. :)

Author: hassony2

README.md

@@ -10,13 +10,13 @@ The idea was to produce the equivalent of [torchvision transforms](https://githu
 
 The basic paradigm is that dataloading should produce videoclips as a **list of PIL Images or numpy.ndarrays** (in format as read by opencv).
 
-Several transforms are then provided in [video_transforms](videotransforms/video_transforms.py).
+Several transforms are then provided in [video_transforms](torchvideotransforms/video_transforms.py).
 
 Each transform iterates on all the images in the list and applies the wanted augmentation.
 
 
 We then have to convert those inputs to torch tensors.
-This can be produced by the [volume_transform](videotransforms/volume_transforms.py).**ClipToTensor** class, which produces a video volume in format (n_channels, n_images, height, width) where n_channels = 3 in case of images.
+This can be produced by the [volume_transform](torchvideotransforms/volume_transforms.py).**ClipToTensor** class, which produces a video volume in format (n_channels, n_images, height, width) where n_channels = 3 in case of images.
 
 When randomness is involved, the same random parameters (crop size, scale size, rotation angle,...) are applied to all the frames.
 
@@ -28,7 +28,7 @@ This should produce something like the top image (this is a dummy clip for now,
 
 # Advancement
 
-[video_transforms.py](videotransforms/video_transforms.py)
+[video_transforms.py](torchvideotransforms/video_transforms.py)
 - [x] Compose
 - [x] Resize
 - [x] CenterCrop
@@ -39,7 +39,7 @@ This should produce something like the top image (this is a dummy clip for now,
 - [x] ColorJitter (acts on brightness, saturation, contrast and hue, only on PIL Images for now)
 - [ ] RandomResizedCrop
 
-[volume_transforms.py](videotransforms/volume_transforms.py)
+[volume_transforms.py](torchvideotransforms/volume_transforms.py)
 - [x] ClipToTensor
 
 
@@ -56,7 +56,7 @@ export PYTHONPATH=$PYTHONPATH:path/to/torch_videovision
 
 ## In your python script
 ```python
-from videotransforms import video_transforms, volume_transforms
+from torchvideotransforms import video_transforms, volume_transforms
 
 video_transform_list = [video_transforms.RandomRotation(30),
 			video_transforms.RandomCrop((200, 200))

setup.py

@@ -0,0 +1,28 @@
+import setuptools
+
+with open("README.md", "r") as fh:
+    long_description = fh.read()
+
+setuptools.setup(
+    name="torchvideotransforms",
+    version="0.1.2",
+    author="Yana Hasson",
+    # author_email="author@example.com",
+    description="A small example package",
+    long_description=long_description,
+    long_description_content_type="text/markdown",
+    url="https://github.com/hassony2/torch_videovision",
+    packages=setuptools.find_packages(),
+    classifiers=[
+        "Programming Language :: Python :: 3",
+        "License :: OSI Approved :: MIT License",
+        "Operating System :: OS Independent",
+        "Development Status :: 2 - Pre-Alpha",
+    ],
+    install_requires=[
+        'torch',
+        'torchvision',
+        'scikit-image',
+    ],
+    python_requires='>=3.6',
+)

testransforms.py

@@ -1,12 +1,10 @@
 import argparse
-import os
 
 from matplotlib import pyplot as plt
-import numpy as np
 from PIL import Image
 
-from videotransforms.video_transforms import Compose, Resize, RandomCrop, RandomRotation, ColorJitter, Normalize
-from videotransforms.volume_transforms import ClipToTensor
+from torchvideotransforms.video_transforms import Compose, Resize, RandomCrop, RandomRotation, ColorJitter, Normalize
+from torchvideotransforms.volume_transforms import ClipToTensor
 
 img_path = 'data/cat/cat1.jpeg'
 
@@ -44,7 +42,7 @@
     RandomCrop(crop_size),
     ColorJitter(args.brightness, args.contrast, args.saturation, args.hue),
     ClipToTensor(channel_nb=channel_nb),
-    Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+    Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
 ]
 video_transform = Compose(video_transform_list)
 

torchvideotransforms/__init__.py

@@ -3,13 +3,12 @@
 import cv2
 import numpy as np
 import PIL
-from skimage.transform import resize
+
 
 def _is_tensor_clip(clip):
     return torch.is_tensor(clip) and clip.ndimension() == 4
 
 
-
 def crop_clip(clip, min_h, min_w, h, w):
     if isinstance(clip[0], np.ndarray):
         cropped = [img[min_h:min_h + h, min_w:min_w + w, :] for img in clip]
@@ -65,7 +64,6 @@ def resize_clip(clip, size, interpolation='bilinear'):
     return scaled
 
 
-
 def get_resize_sizes(im_h, im_w, size):
     if im_w < im_h:
         ow = size
@@ -75,6 +73,7 @@ def get_resize_sizes(im_h, im_w, size):
         ow = int(size * im_w / im_h)
     return oh, ow
 
+
 def normalize(clip, mean, std, inplace=False):
     if not _is_tensor_clip(clip):
         raise TypeError('tensor is not a torch clip.')

videotransforms/functional.py torchvideotransforms/functional.py

@@ -2,7 +2,7 @@
 import PIL
 import torch
 
-from videotransforms.utils import images as imageutils
+from .utils import images as imageutils
 
 
 class ToStackedTensor(object):

videotransforms/stack_transforms.py torchvideotransforms/stack_transforms.py

@@ -1,6 +1,6 @@
 import random
 
-from videotransforms.utils import functional as F
+from .utils import functional as F
 
 
 class Normalize(object):

videotransforms/tensor_transforms.py torchvideotransforms/tensor_transforms.py

@@ -1,12 +1,8 @@
 import numbers
 import random
-
-import cv2
-from matplotlib import pyplot as plt
 import numpy as np
 import PIL
 import skimage
-import torch
 import torchvision
 
 from . import functional as F

torchvideotransforms/utils/__init__.py

@@ -2,7 +2,7 @@
 from PIL import Image
 import torch
 
-from videotransforms.utils import images as imageutils
+from .utils import images as imageutils
 
 
 class ClipToTensor(object):