aleju / imgaug Goto Github PK

First of all thank you very much for this.
When I try to run your example code with these two images, I always get a black image!

This is the whole code which is given in the first page , I just replaced the random numpy array statement with these two images!:

import imgaug as ia
from imgaug import augmenters as iaa
import numpy as np

im = caffe.io.load_image('buckskin_s_000331.png')
im2 = caffe.io.load_image('buckskin_s_000005.png')
images = np.zeros([2,32,32,3])
images[0] = im
images[1] = im2
# Sometimes(0.5, ...) applies the given augmenter in 50% of all cases,
# e.g. Sometimes(0.5, GaussianBlur(0.3)) would blur roughly every second image.
st = lambda aug: iaa.Sometimes(0.3, aug)

# Define our sequence of augmentation steps that will be applied to every image
# All augmenters with per_channel=0.5 will sample one value _per image_
# in 50% of all cases. In all other cases they will sample new values
# _per channel_.
seq = iaa.Sequential([
        iaa.Fliplr(0.5), # horizontally flip 50% of all images
        iaa.Flipud(0.5), # vertically flip 50% of all images
        st(iaa.Superpixels(p_replace=(0, 1.0), n_segments=(20, 200))), # convert images into their superpixel representation
        st(iaa.Crop(percent=(0, 0.1))), # crop images by 0-10% of their height/width
        st(iaa.GaussianBlur((0, 3.0))), # blur images with a sigma between 0 and 3.0
        st(iaa.Sharpen(alpha=(0, 1.0), strength=(0.75, 1.5))), # sharpen images
        st(iaa.Emboss(alpha=(0, 1.0), strength=(0, 2.0))), # emboss images
        # search either for all edges or for directed edges
        st(iaa.Sometimes(0.5,
            iaa.EdgeDetect(alpha=(0, 0.7)),
            iaa.DirectedEdgeDetect(alpha=(0, 0.7), direction=(0.0, 1.0)),
        )),
        st(iaa.AdditiveGaussianNoise(loc=0, scale=(0.0, 0.2), per_channel=0.5)), # add gaussian noise to images
        st(iaa.Dropout((0.0, 0.1), per_channel=0.5)), # randomly remove up to 10% of the pixels
        st(iaa.Invert(0.25, per_channel=True)), # invert color channels
        st(iaa.Add((-10, 10), per_channel=0.5)), # change brightness of images (by -10 to 10 of original value)
        st(iaa.Multiply((0.5, 1.5), per_channel=0.5)), # change brightness of images (50-150% of original value)
        st(iaa.ContrastNormalization((0.5, 2.0), per_channel=0.5)), # improve or worsen the contrast
        st(iaa.Affine(
            scale={"x": (0.8, 1.2), "y": (0.8, 1.2)}, # scale images to 80-120% of their size, individually per axis
            translate_px={"x": (-16, 16), "y": (-16, 16)}, # translate by -16 to +16 pixels (per axis)
            rotate=(-45, 45), # rotate by -45 to +45 degrees
            shear=(-16, 16), # shear by -16 to +16 degrees
            order=ia.ALL, # use any of scikit-image's interpolation methods
            cval=(0, 255), # if mode is constant, use a cval between 0 and 255
            mode=ia.ALL # use any of scikit-image's warping modes (see 2nd image from the top for examples)
        )),
        st(iaa.ElasticTransformation(alpha=(0.5, 3.5), sigma=0.25)) # apply elastic transformations with random strengths
    ],
    random_order=True # do all of the above in random order
)

images_aug = seq.augment_images(images)
plt.imshow(images_aug[0])
plt.show()

what is wrong here?

Hi Alex! This library is amazing!! However when running the code below I get this error: RuntimeError: Could not execute image viewer.

import augmenters as iaa
import numpy as np

images = np.random.randint(0, 255, (16, 128, 128, 3), dtype=np.uint8)
seq = iaa.Sequential([iaa.Fliplr(0.5), iaa.GaussianBlur((0, 3.0))])

seq.show_grid(images[0], cols=8, rows=8)

seq.show_grid([images[0], images[1]], cols=8, rows=8)

I met some errors in the use of the library. when I launch my anaconda environment and come into Python script,the following line works well:
from imgaug import augmenters as iaa
but when I come into Jupyter book rather than Python script, the same line will get an error.
why is it?

I know there has to be a limit to asking questions.
But I am working on a new feature learning methodology.

My algorithm has learned this following set of filters on MNIST. I was hoping you could give me a second opinion over how good these filters are.

features_1.pdf
features_2.pdf

Also, which ones of the two seem better?

Could you publish this package to PyPi? It would make installing (e.g. pip install imgaug) and using it in other packages more easy. Thanks!

after load image,make draw_grid(image,cols=8,rows=8),there have a problem boundary mode not supported.how to deal with?

As said in the README, I copied the required files into my directory and ran the
from imgaug import augmenters as iaa

but I get an error

"ImportError: cannot import name 'augmenters'"

How does the Affine Transformation works?
What I mean to ask is, out of the selected batch, upon transforming the image batch with the transformer,
are there non-transformed images in the resulting batch or has every sample obtained certainly gone under some transformation?
LIke, I know the Flip transformers only transform specified ratio of images. What about other transformers?

Four augmenters that use the Convolve class: (Sharpen, Emboss, EdgeDetect and DirectEdgeDetect) are giving me pickle errors.

AttributeError: Can't pickle local object 'Emboss.<locals>.create_matrices'
Any easy way to fix this or make them pickle safe?

Regards @aleju

This looks amazing!! How to install and make use of this library?

Not a bug, a question.

Is it possible to use the crop function to randomly crop an image to a fixed size? I.e.: crop 256x256 images to 224x224 at a random position (within the image area)?

Thank you.

Hi, when I try to import the augmenters file from imgaug root folder using iPython I get this error
ImportError: cannot import name augmenters

Also, when I do the import import augmenters as iaa from the imgaug/imgaug folder, I get this error

       1 from __future__ import print_function, division, absolute_import
----> 2 from . import imgaug as ia
      3 from .parameters import StochasticParameter, Deterministic, Binomial, Choice, DiscreteUniform, Normal, Uniform
      4 from abc import ABCMeta, abstractmethod
      5 import random

ValueError: Attempted relative import in non-package

However, when importing from a python script it works fine

I'm getting this error:

Import Error: cannot import name augmenters

Whenever I try to:
from imgaug import augmenters as iaa
-I'm running windows 7
-Anaconda with python 2
-Installed imgaug using this line:
pip install git+https://github.com/aleju/imgaug
-Running the script using visual studio 2015

Hi!

Thanks a lot for the library, it really works well. The only thing I have issues with is the keypoint augmentation, which works sometimes but in other cases it returns a set of negative coordinate values which are then of course not plotted inside the image when I use 'draw_on_image' and show it using matplotlib afterwards.
I'm currently using this augmentation sequence:

seq = iaa.Sequential([
    iaa.Fliplr(0.5), # horizontally flip 50% of the images
    iaa.Flipud(0.5), # vertically flip 50% of all images
    #iaa.GaussianBlur(sigma=(0, 3.0)), # blur images with a sigma of 0 to 3.0
    iaa.Affine(
            scale={"x": (0.8, 1.2), "y": (0.8, 1.2)}, # scale images to 80-120% of their size, individually per axis
            translate_px={"x": (-16, 16), "y": (-16, 16)}, # translate by -16 to +16 pixels (per axis)
            rotate=(-45, 45), # rotate by -45 to +45 degrees
            shear=(-16, 16), # shear by -16 to +16 degrees
            order=ia.ALL, # use any of scikit-image's interpolation methods
            #cval=(0, 1.0), # if mode is constant, use a cval between 0 and 1.0
            #mode=ia.ALL # use any of scikit-image's warping modes (see 2nd image from the top for examples)
        )
    ],
    random_order=True # do all of the above in random order
)

seq_det = seq.to_deterministic() # call this for each batch again, NOT only once at the start

For example, when I use this input image:

I get results like this:

Does anyone have an idea what could be the problem?

Thanks a lot!

UPDATE: It seems like the keypoint augmentation only works if a single command is used in iaa.Affine() at once. If use several, only the first is use to augment keypoints. What works is to use several iaa.Affine() commands in series, with each containing a different image manipulation, which is fine for now.

it's more of a question rather than an issue.
Have you considered implementing ZCA whitening?
I have a simple implementation. But It doesn't seem to work on color images.

For example, on CIFAR10 dataset, the result of ZCA whitening is distorted images where only few color pixels remain and rest of the image gets all distorted.

Also, ZCA whitening acts as a low pass filter according to some sources. What are other techniques that act as a low pass filter? Gaussian Blurring???

Hello!
This is not an issue, but instead a review of the installation process.

The installation script has successfully worked for

1. virtualenvs
2. global installs
3. conda and anaconda

tested with both python 2.7 and python3.5

This was good work !!! 👍

In the example found here, the first example references some load_batches command that's not defined earlier. Searching in the augmenters.py file (imported as iaa) also did not yield any results).

What is that method?

Hi,
Imgaug is a very good tool for generating images. I am trying to generate different data from my 3D MRI image data. As the code is only supporting 2D images or RGB images, imgaug is giving failure in case of my data. This might seem as my personal different problem, but I want your suggestion of how can we change the code so that it will support 3D images. My images are 25625660 sizes. Its a dicom image, and it has 60 slices. Generating augmented 2D image will not touch Z axis, which is not acceptable in my case. Please suggest something using which I can achieve the same feature, mainly I will use affine,piecewise affine and elastic transformation for generating new data.

Thanks

I'd like to augument input image recording of S and V channels of HSV image.
But in my knowledge Add class does not have parameter for specifying channel, and it is not possible currently.
https://github.com/aleju/imgaug/blob/master/imgaug/augmenters.py#L1551
Is this possible in actual, and am I missing something?

I came across this error just now. When loading the augmenters module in Jupyter notebooks, the following line works

import augmenters as iaa

However, the same line fails if used in a script (in the same directory).
Any Idea what might be happening?? I think it's the (rather notorious) relative import system of python that's to blame.

PS: I forgot to tell you. I've been testing all of it in Python 3.5, and the import troubles (earlier and now this) are the only problems I have encountered.

Hi, I've got the following code:

def augment(im, y):
    im_arr = np.array(im)

    # See documentation for details regarding transformations: https://github.com/aleju/imgaug
    fliplr_rate = 0.5
    angle = 10
    additive, contrast_norm = (45, 0.1)
    gaussian_noise, dropout = (0.05, 0.01)
    shear, shift = (2, 20)
    aug_img_only = iaa.Sequential([
        iaa.Sometimes(0.5, iaa.OneOf([
            iaa.Add((-additive, additive)),
            iaa.ContrastNormalization((1 - contrast_norm, 1 + contrast_norm))
        ])),
        iaa.Sometimes(0.5, iaa.OneOf([
            iaa.AdditiveGaussianNoise(scale=gaussian_noise * 255, per_channel=True),
            iaa.Dropout(dropout)
        ]))
    ])
    aug_img_mask = iaa.Sequential([
        iaa.Fliplr(fliplr_rate),
        iaa.Affine(rotate=(-angle, angle)),
        iaa.Sometimes(0.5, iaa.Affine(
            shear=(-shear, shear),
            translate_px={'x': (-shift, shift), 'y': (-shift, shift)})
                      )
    ])
    aug_img_only.reseed()
    aug_img_only_det, aug_img_mask_det = aug_img_only.to_deterministic(), aug_img_mask.to_deterministic()
    im_arr = aug_img_only_det.augment_images([im_arr])[0]
    im_arr = aug_img_mask_det.augment_images([im_arr])[0]
    y = aug_img_mask_det.augment_images([y])[0]

    im = Image.fromarray(im_arr)
    return im, y

I've got a ML system which has input images and known masks of areas of interest, which I later want to predict. I want to augment the images and the masks in the same way for some transformations, and apply other transformations (such as dropout, etc.) only to the original image.
Here, in the code, im is the original image in PIL object format, im_arr is the original image transformed to numpy array, and y is the mask numpy array.
Now, everytime I run this code, for example, 5 times, with the same picture and mask, I get the same 5 augmentations. Meaning, that the first picture comes out the same every time, so does the second and so on.
Just to clarify, here is the code I use to run it:

for i in range(5):
    im = Image.open('image.jpg')
    y = np.load('mask.npy')
    im, y = augment(im, y)

Why would this behavior happen? I reinstantiate the augmenters every time the function is called (as can be seen in the code), and only after the reinstantiation do I call to_deterministic().
What am I missing?

Thanks in advance!

Can you tell me the meaning of this parameter?? It's almost everywhere

Also. Keypoints_on_images seems to be a list object. Are individual keypoints list objects themselves? Or numpy array?

I notice that in the example of README.md, all examples related with augmenting landmark points have a following line of code. However, it seems that the examples that only augment images do not.
seq_det = seq.to_deterministic() # call this for each batch again, NOT only once at the start
Is it necessary only for landmarks augmentation?

when I load a image , do some augmentations ,the result is abnormal with lots of strides
`
seq = iaa.Sequential([
#iaa.Flipud(0.5),

#iaa.Dropout(0.02, name="Dropout"),

#iaa.GaussianBlur(5)

#iaa.AdditiveGaussianNoise(0, 10, True)

iaa.Sharpen(alpha=(0, 1.0), lightness=(0.75, 1.5))

#iaa.Crop(px=(0, 16)), # crop images from each side by 0 to 16px (randomly chosen)

#iaa.Fliplr(0.5), # horizontally flip 50% of the images

#iaa.GaussianBlur(sigma=(0, 5.0)) # blur images with a sigma of 0 to 3.0

])
img = scipy.misc.imread(image_dir)
images_aug = seq.augment_images(img)
misc.imshow(images_aug)
`

Currently the `generate_example_images.py' script only generates an ' examples_grid.jpg' image. May I ask that how to save every sample in this grid image into separate image?
Thanks!

Hi, thanks for writing this repo! It's a great help in what I do :)

One question:
When I scale the image by 1.5 (affine transform) I lose information in the sides (it's more of a zoom than a resize function). Is there any way to simply resize the image?

Similar examples would be:

• When I use affine to scale the image by 0.5, I'm getting an image with a lot of black and have to perform a crop later
• When I rotate an image, the corners become cropped in the result.

The solution I am looking for is something like adding a flag "canvas_fits_image" which will perform the affine transformation and resize the canvas accordingly.

If it doesn't exist, I might add it myself sometime. Thanks!

Hello,

l passed this code generate_example_images.py (in a for loop ) on a dataset of 200.000 images. After that l noticed that not all the images are augmented . For instance , additivegaussian noise for some images it worked perfectly but for other they remained the same. This also apply for other data augmentation method.
What is wrong ?

Thank you

Hi,

I am using sequential augment with only rotate affine transform with -30, 30 range.

Then I wanted to augment keypoints. I did this with the point (100, 100). But, the augmented point is not in the correct position. I ran the key point augmentation for other sequential non-affine augmentations and they seemed to work fine.

I used the same visualization technique in the README.md with the "draw_on_image" method

Can you please help me with this.

When I was running "from imgaug import augmenters as iaa"
It has an error:
Intel MKL FATAL ERROR: Cannot load libmkl_avx2.so or libmkl_def.so.

But the libmkl_avx2.so and libmkl_def.so are in the folder ~/anaconda2/lib/.

I run "sudo -H python -m pip install imgaug-0.2.4.tar.gz" command to user my pip in Cellar to install imgaug, and got the following error.
And i tried "pip install imgaug-0.2.4.tar.gz", and it seems that it uses 2.7.10 version python in osx system. However, my opencv is installed in 2.7.11 python in Cellar folder.
How can i solve this problem?

Processing ./imgaug-0.2.4.tar.gz
Error [Errno 2] No such file or directory while executing command python setup.py egg_info
Exception:
Traceback (most recent call last):
File "/usr/local/lib/python2.7/site-packages/pip/basecommand.py", line 215, in main
status = self.run(options, args)
File "/usr/local/lib/python2.7/site-packages/pip/commands/install.py", line 335, in run
wb.build(autobuilding=True)
File "/usr/local/lib/python2.7/site-packages/pip/wheel.py", line 749, in build
self.requirement_set.prepare_files(self.finder)
File "/usr/local/lib/python2.7/site-packages/pip/req/req_set.py", line 380, in prepare_files
ignore_dependencies=self.ignore_dependencies))
File "/usr/local/lib/python2.7/site-packages/pip/req/req_set.py", line 634, in _prepare_file
abstract_dist.prep_for_dist()
File "/usr/local/lib/python2.7/site-packages/pip/req/req_set.py", line 129, in prep_for_dist
self.req_to_install.run_egg_info()
File "/usr/local/lib/python2.7/site-packages/pip/req/req_install.py", line 439, in run_egg_info
command_desc='python setup.py egg_info')
File "/usr/local/lib/python2.7/site-packages/pip/utils/init.py", line 667, in call_subprocess
cwd=cwd, env=env)
File "/usr/local/Cellar/python/2.7.11/Frameworks/Python.framework/Versions/2.7/lib/python2.7/subprocess.py", line 710, in init
errread, errwrite)
File "/usr/local/Cellar/python/2.7.11/Frameworks/Python.framework/Versions/2.7/lib/python2.7/subprocess.py", line 1335, in _execute_child
raise child_exception
OSError: [Errno 2] No such file or directory

Hi!

I think the case of augmentation by perspective transform is very useful for real-life applications. Would you be interested in such function?

See http://www.euclideanspace.com/maths/discrete/groups/categorise/finite/cube/rotFace.png . Let the plane with a green 1 be an image. I'd suggest to parametrize the augmentation by angles around z and y, i.e. img_aug = aug_by_perspective(img, angle_vert, angle_horiz).

Implementing this should be relatively simple using OpenCV.

Hello,

I am running the example code on my set of images and it's working great. But I want to save the resulted images one by one rather than saving them in a grid.

How can we convert these lines of code to save the images one by one rather than a grid?

grid = seq.draw_grid(image, cols=8, rows=8)
misc.imsave("examples_grid.jpg", grid)

Thanks

I have added (some) documentation for the Augmenter and Noop classes in the augmenters.py file.

Review the changes in My fork and make suggestions.

This is just a glimpse. Much work will be needed on documentation (as more often than not I am not quite sure what a parameter signifies, etc).

I found the "Sequential model" in examples of this library. I learn that the model belongs to Keras. Whether it means that I need to install Keras in addition ?

May you add the script to generate ' examples_grid.jpg' image?
Thanks!

Segmentation labels (annotation) are usually provided as a matrix of labels (integers).
For this task the same augmentation as the images should be performed, except interpolation should be 'nearest neighbours' rather than bilinear or similar one.
Some form of retrieving the sampled parameters (for example: rotation angle was 32 degrees) would be usefull so we could re-create the same augmentation, but with different interpolation.
Deterministic augmentation doesn't help in this case.
In the meantime, as a workaround we can use:

from imgaug import augmenters as iaa
rotate_max=30
seq = iaa.Sequential([
    iaa.Fliplr(0.5), # horizontally flip 50% of the images
],random_state=0)
rotate=iaa.Affine(rotate=(0, rotate_max),mode='edge')
rotate_nn_interpolation=iaa.Affine(rotate=(0, rotate_max),order=[0],mode='edge')
seq_nn_interpolation=seq.deepcopy()
seq.append(rotate)
seq_nn_interpolation.append(rotate_nn_interpolation)

aug_seed=0
for batch in...:
    batch_images = images[batch, :, :, :]
    batch_labels = labels[batch, :,:]
    seq.reseed(aug_seed)
    seq_nn_interpolation.reseed(aug_seed)
    batch_images=seq.augment_images(batch_images)
    batch_labels=seq_nn_interpolation.augment_images(batch_labels)
    aug_seed+=1

When I clone the repo, and run the generate_example_images.py, I get a runtime error:

$ cd ~/repos/imgaug
$ python generate_example_images.py
[draw_per_augmenter_images] Loading image...
[draw_per_augmenter_images] Initializing...
[draw_per_augmenter_images] Augmenting...
Traceback (most recent call last):
  File "generate_example_images.py", line 290, in <module>
    main()
  File "generate_example_images.py", line 18, in main
    draw_per_augmenter_images()
  File "generate_example_images.py", line 252, in draw_per_augmenter_images
    misc.imsave("examples.jpg", output_image.draw())
  File "generate_example_images.py", line 271, in draw
    rows_drawn = [self.draw_row(title, images, subtitles) for title, images, subtitles in self.rows]
  File "generate_example_images.py", line 277, in draw_row
    title_cell = ia.draw_text(title_cell, x=2, y=2, text=title, color=[0, 0, 0], size=12)
  File "/Users/erickim/repos/imgaug/imgaug/imgaug.py", line 129, in draw_text
    font = ImageFont.truetype("DejaVuSans.ttf", size)
  File "/usr/local/lib/python2.7/site-packages/PIL/ImageFont.py", line 238, in truetype
    return FreeTypeFont(font, size, index, encoding)
  File "/usr/local/lib/python2.7/site-packages/PIL/ImageFont.py", line 127, in __init__
    self.font = core.getfont(font, size, index, encoding)
IOError: cannot open resource

A quick fix is to modify imgaug/imgaug.py:128 and give the absolute path of the DejaVuSans.ttf file that is included in the repo:

diff --git a/imgaug/imgaug.py b/imgaug/imgaug.py
index 7e94c82..b2b2485 100644
--- a/imgaug/imgaug.py
+++ b/imgaug/imgaug.py
@@ -9,6 +9,7 @@ import math
 from scipy import misc
 import multiprocessing
 import threading
+import os
 import sys
 import six
 import six.moves as sm
@@ -125,7 +126,8 @@ def draw_text(img, y, x, text, color=[0, 255, 0], size=25):

     shape = img.shape
     img = Image.fromarray(img)
-    font = ImageFont.truetype("DejaVuSans.ttf", size)
+    font = ImageFont.truetype(os.path.join(os.path.abspath(os.path.split(__file__)[0]), "DejaVuSans.ttf"), size)
+
     context = ImageDraw.Draw(img)
     context.text((x, y), text, fill=tuple(color), font=font)
     img_np = np.asarray(img)

Thoughts on this change?

I am using python3 without conda/env.

After running:
sudo pip3 install git+https://github.com/aleju/imgaug
I receive:

Collecting git+https://github.com/aleju/imgaug
  Cloning https://github.com/aleju/imgaug to /tmp/pip-mn453yqe-build
*** Error in `/usr/bin/python3': free(): invalid pointer: 0x0000000001af2990 ***
======= Backtrace: =========
...
======= Memory map: ========
...
Complete output from command python setup.py egg_info:
    running egg_info
    creating pip-egg-info/imgaug.egg-info
    writing top-level names to pip-egg-info/imgaug.egg-info/top_level.txt
    writing requirements to pip-egg-info/imgaug.egg-info/requires.txt
    writing pip-egg-info/imgaug.egg-info/PKG-INFO
    writing dependency_links to pip-egg-info/imgaug.egg-info/dependency_links.txt
    writing manifest file 'pip-egg-info/imgaug.egg-info/SOURCES.txt'
    reading manifest file 'pip-egg-info/imgaug.egg-info/SOURCES.txt'
    writing manifest file 'pip-egg-info/imgaug.egg-info/SOURCES.txt'
    
    ----------------------------------------
Command "python setup.py egg_info" failed with error code -6 in /tmp/pip-mn453yqe-build/

I am using manually build OpenCV 3.1.0-dev (from pyimagesearch tutorial)

When run the setup, error happens. opencv is installed on Anaconda.
Is it possible to install imgaug on Anaconda?
...
Processing ./dist/imgaug-0.2.0.tar.gz
Complete output from command python setup.py egg_info:
Traceback (most recent call last):
File "", line 1, in
File "/tmp/pip-K5MRPU-build/setup.py", line 6, in
raise Exception("Could not find package 'cv2' (OpenCV). It cannot be automatically installed, so you will have to manually install it.")
Exception: Could not find package 'cv2' (OpenCV). It cannot be automatically installed, so you will have to manually install it.

As far as I can see your rotation is intended for square images. I am interested in using rotation like in the following code:

from scipy.ndimage.interpolation import rotate
# some code
new_image = rotate(image, angle, reshape=True)

Thus, it will transform image with shape (h,w) into image with shape (w,h) without any scaling by one of the axies. Can I make it using your library in some way or should I make a pull request?

Hi,

Is it difficult to generalize image augmentations without the restrictions of np.uint8 and 0-255 ?
I saw in the code that there are some augmentations that have a hard-coded np.clip and astype np.uint8.
I started locally to modify test.py to launch tests on float32 data, however tests are not exaustive...
I'm interested in this feature and can help with implementation.

Suppose I have video-frames and I would like to apply same distortions/transformation/augmentations to all the frames while passing them in a single batch(and then other batch with other parameters for the same transformation set). How can this be done?
I am sorry if my query seems to be too naive

@aleju Thank you for providing such an amazing package which makes convenient for training deep networks for visual recognition. In the example codes,

# Sometimes(0.5, ...) applies the given augmenter in 50% of all cases,
# e.g. Sometimes(0.5, GaussianBlur(0.3)) would blur roughly every second image.
st = lambda aug: iaa.Sometimes(0.5, aug)

I am still confused by the functionality of Sometimes(0.5), would you please give more comments on this function? Thank you .

In the example of "heavy augmentations" I get the error

order=ia.ALL, # use any of scikit-image's interpolation methods
AttributeError: 'module' object has no attribute 'ALL'

I fixed this by replacing ia.ALL with "ALL"
Does the variable ALL accept other values?

PS: Great library. Very good work

I am not sure how to ask this question properly so bare with me.

I am going to try to use an illustrative example:

1. Consider a door on its hinges. Lets say door opens away from you (so that you have to push the door open rather than pull it towards you). Picture the door rotating on its hinge away from you, the door nob now appears further away from you, assuming you pushed the door and stayed in the same spot.

Is there an image translation that does the equivalent of this? So the "door knob" portions of the image would appear further away and the portions of the image closest to the "hinge" would appear closer.

Similar idea to these?

I am not looking for a 3d images I am just wanting some portions of my image closer and some portions further away.

Hello,

l have a set of images based sequence. l want to increase my dataset by doing some data augmentation as follow. for each image add horizontal line, vertical line at on the most left char and most right char and in the middle.

Here is the original image :

and l want to get something like this :

Adding horizontal line :

Adding vertical line on the middle

Adding vertical line on the left

Adding vertical line on the right

thank you

Can you tell me which operation is slow so i can remove it. Thx!

here is my code

seq = iaa.Sequential([
        iaa.Fliplr(0.5), 
        iaa.Flipud(0.5),
        st(iaa.GaussianBlur((0, 1.0))), 
        st(iaa.Sharpen(alpha=(0, 1.0), lightness=(0.75, 1.5))), 
        st(iaa.Emboss(alpha=(0, 1.0), strength=(0, 2.0))),
        st(iaa.Sometimes(0.5,
            iaa.EdgeDetect(alpha=(0, 0.4)),
            iaa.DirectedEdgeDetect(alpha=(0, 0.4), direction=(0.0, 1.0)),
        )),
        st(iaa.AdditiveGaussianNoise(loc=0, scale=(0.0, 0.2), per_channel=0.5)), 
        st(iaa.Dropout((0.0, 0.1), per_channel=0.5)), 
        st(iaa.Add((-10, 10), per_channel=0.5)), 
        st(iaa.Multiply((0.5, 1.5), per_channel=0.5)),
        st(iaa.ContrastNormalization((0.5, 1.5), per_channel=0.5)),
        st(iaa.ElasticTransformation(alpha=(0.5, 2.5), sigma=0.25))
    ],
    random_order=True 
)

Hi, Thanks for the library!

When we do the rotation transformation, can we support 'edge' padding for the black areas? skimage supports different padding mode: http://scikit-image.org/docs/dev/api/skimage.transform.html#skimage.transform.rotate

Thank you!

[Enhancement] It would be nice if cval was added to the __init__ function as a parameter with a description. I understand you can set it manually at the moment but it takes a while to even see if the code supports setting the background color.