Streaming over lightweight data transformations

Code to reproduce:

import solt
import solt.transforms as slt
import numpy as np


if __name__ == "__main__":
    trf = solt.Stream([slt.Resize(resize_to=(50, 50)),
                       slt.Flip()])
    img1 = np.ones((100, 100, 3), dtype=np.int32)
    img2 = np.ones((110, 110, 3), dtype=np.int32)
    trf_img = trf({'images': (img1, img2)})
    print('Done')

Adding a resize transformation at the beginning doesn't help. A clear error message is needed for this case.

While testing one of the notebook algorithm, this error occured: kpts = None for annotation_fname in glob.glob(os.path.join('Data', 'helen_annotations', '*.txt')): with open(annotation_fname) as f: if f.readline()[:-1] == fname.split('.')[0]: ktps = [] for l in f: tmp = l.split() ktps.append([float(tmp[0]), float(tmp[2])]) break kpts = np.array(ktps)

Specific Error:

NameError Traceback (most recent call last) in 8 ktps.append([float(tmp[0]), float(tmp[2])]) 9 break ---> 10 kpts = np.array(ktps)

NameError: name 'ktps' is not defined

Proof of concept.

• Switched to pytorch for 3D+ in Flip, Crop and Pad
• Flip now accepts multiple axes, but not axis=-1. Also, the axis order has been switched from opencv to numpy
• Minor cleanup to the tests

• Removed height and width concepts where possible
• Switched from shape checking decorator to function
• Added shape validation also for masks and keypoints
• Splitted utils into checks and serial

Bumps opencv-python-headless from 4.1.2.30 to 4.2.0.32.

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

https://github.com/MIPT-Oulu/solt/blob/485bfb0d471134f75e09d06aa5e9ce4b57c0e13c/solt/transforms/_transforms.py#L1018 tells something about noise. Should talk about the intensities instead.

Works fine with 1.1, but not 1.5. Traceback:

    as_dict=as_dict, scale_keypoints=scale_keypoints, normalize=normalize, mean=mean, std=std,
  File ".../site-packages/solt/core/_data.py", line 273, in to_torch
    img.sub_(mean)
RuntimeError: output with backend CPU and dtype Byte doesn't match the desired backend CPU and dtype Float
Process finished with exit code 1

🐛 Bug

It is not possible to run unit tests locally on MacOS X (10.12).

To Reproduce

Steps to reproduce the behavior:

1. Create a conda environment and install all of the packages as suggested here.

2. Run the following code: pytest tests

3. The unit tests fail and generate an error:

ImportError while importing test module '/Users/melekhi1/projects/solt/tests/test_data_core.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
tests/test_data_core.py:4: in <module>
    import cv2
../../anaconda2/envs/solt_test_env/lib/python3.8/site-packages/cv2/__init__.py:3: in <module>
    from .cv2 import *
E   ImportError: dlopen(/Users/melekhi1/anaconda2/envs/solt_test_env/lib/python3.8/site-packages/cv2/cv2.cpython-38-darwin.so, 2): Symbol not found: _inflateValidate
E     Referenced from: /Users/melekhi1/anaconda2/envs/solt_test_env/lib/python3.8/site-packages/cv2/.dylibs/libpng16.16.dylib (which was built for Mac OS X 10.13)
E     Expected in: /usr/lib/libz.1.dylib
E    in /Users/melekhi1/anaconda2/envs/solt_test_env/lib/python3.8/site-packages/cv2/.dylibs/libpng16.16.dylib

Expected behavior

The unit tests should pass

=============================== test session starts ================================
platform darwin -- Python 3.8.2, pytest-3.6.4, py-1.8.1, pluggy-0.7.1
rootdir: /Users/melekhi1/projects/solt, inifile:
plugins: pep8-1.0.6, flake8-1.0.2, cov-2.6.0
collected 701 items

tests/test_base_transforms.py .............................................. [  6%]
............................................................................ [ 17%]
.....                                                                        [ 18%]
tests/test_data_core.py .................................................... [ 25%]
............................................................................ [ 36%]
............................................................................ [ 47%]
..................................................................           [ 56%]
tests/test_transforms.py ................................................... [ 63%]
............................................................................ [ 74%]
............................................................................ [ 85%]
............................................................................ [ 96%]
.........                                                                    [ 97%]
tests/test_utils.py ................                                         [100%]

============================ 701 passed in 6.79 seconds ============================

Suggestions

It seems the bug is related to the latest version of opencv-python-headless (4.2.0.32) package. Apparently, its previous version, 4.1.2.30, could solve the issue at least for this particular platform (MacOS 10.12). It would be great to test this solution on more recent MacOS.

The attribute crop_size needs to be renamed into crop_to to match the constructor. Otherwise, the serialization does not work well.

https://github.com/MIPT-Oulu/solt/blob/770e397884bcafe80a11723c229e275c1c1f8b5a/solt/transforms/_transforms.py#L715

the augbench package needs further development and its results need to be reported in the README.

the benchmark needs to cover random transformations rather than the static ones. It is important to make comparison for three cases: image, image mask, and image+10 masks (instance segmentation task).

The use case for this is that for some images it might be better to rely on their mean intensity and increase their intensity by n% of the mean value.

Currently, selective stream takes a parameter that tells how many transforms to select from the list of transformations. I was recently suggested to make it a range and allow to sample random number of augmentations out of this range.