Fill color support for tensor affine transforms #2904

voldemortX · 2020-10-27T13:01:06Z

PR for Issue #2887

It seems that affine(), rotate() and perspective() all use _apply_grid_transform(), so I added the code there.
I ran unit tests for all 3 functions and passed, but I can't figure out whether it is right for bilinear. e.g. If I change

vision/test/test_functional_tensor.py

Line 553 in f9e31a6

for r in [0, ]:

to [0, 2], both the original code and the current commit can't pass the tests.

p.s. I only have CUDA 10.0 at the moment, so I had to block the version check in torchvision/extensions.py, not sure if that caused the above mentioned test failures.

vfdev-5

Thanks a lot for the PR @voldemortX !
I left few comments for minor improvements. We have to update also the docstring for F.affine and others.

torchvision/transforms/functional_tensor.py

vfdev-5 · 2020-10-27T14:12:57Z

I can't figure out whether it is right for bilinear

@voldemortX this is a good question and you are right, those tests do not pass as interpolation results are different between PIL and torch. So, for instance, we can skip the tests for resample=2.

voldemortX · 2020-10-27T14:19:10Z

Actually I have another question needs clearing: If all we are considering is linear operations (Euclidean transforms and linear interpolations), the dummy mask should always be 0/1 after affine ops? Although even if that is the case I'd still prefer sticking with 0.5 thresholding.

vfdev-5 · 2020-10-27T14:28:55Z

the dummy mask should always be 0/1 after affine ops?

Actually, no if resample>0. You'll have float values on the boundaries due to interpolations between 0 and 1, right? For example

import torch
from torchvision.transforms.functional import affine

mask = torch.ones(1, 8, 8, dtype=torch.float32)
res = affine(mask, angle=45, translate=[0.1, 0.0], scale=1.0, shear=[0., 0.], resample=2)
res[0, :, :]
> 
tensor([[0.0000, 0.1866, 0.8938, 1.0000, 1.0000, 1.0000, 0.3281, 0.0000],
        [0.1866, 0.8938, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 0.3281],
        [0.8938, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [0.8938, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [0.1866, 0.8938, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 0.3281],
        [0.0000, 0.1866, 0.8938, 1.0000, 1.0000, 1.0000, 0.3281, 0.0000]])

voldemortX · 2020-10-27T14:45:52Z

the dummy mask should always be 0/1 after affine ops?

Actually, no if resample>0. You'll have float values on the boundaries due to interpolations between 0 and 1, right? For example

import torch
from torchvision.transforms.functional import affine

mask = torch.ones(1, 8, 8, dtype=torch.float32)
res = affine(mask, angle=45, translate=[0.1, 0.0], scale=1.0, shear=[0., 0.], resample=2)
res[0, :, :]
> 
tensor([[0.0000, 0.1866, 0.8938, 1.0000, 1.0000, 1.0000, 0.3281, 0.0000],
        [0.1866, 0.8938, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 0.3281],
        [0.8938, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [0.8938, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [0.1866, 0.8938, 1.0000, 1.0000, 1.0000, 1.0000, 1.0000, 0.3281],
        [0.0000, 0.1866, 0.8938, 1.0000, 1.0000, 1.0000, 0.3281, 0.0000]])

Well in that case, if we replace a pixel's value with the fillcolor directly in bilinear resample mode (like what just implemented), it would be a different behavior than this?

Maybe we need img * mask + (1 - mask) * fillcolor instead?

codecov · 2020-10-27T17:01:25Z

Codecov Report

Merging #2904 into master will decrease coverage by 0.08%.
The diff coverage is 80.48%.

@@            Coverage Diff             @@
##           master    #2904      +/-   ##
==========================================
- Coverage   73.41%   73.32%   -0.09%     
==========================================
  Files          99       99              
  Lines        8801     8840      +39     
  Branches     1389     1397       +8     
==========================================
+ Hits         6461     6482      +21     
- Misses       1915     1930      +15     
- Partials      425      428       +3

Impacted Files	Coverage Δ
torchvision/transforms/functional.py	`80.44% <ø> (ø)`
torchvision/transforms/functional_tensor.py	`73.83% <75.00%> (-0.32%)`	⬇️
torchvision/transforms/transforms.py	`80.51% <84.00%> (-0.01%)`	⬇️
torchvision/io/__init__.py	`68.75% <0.00%> (-20.91%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f9e31a6...50d311d. Read the comment docs.

vfdev-5 · 2020-10-27T17:04:10Z

@voldemortX I see what you mean about img * mask + (1 - mask) * fillcolor. Probably, this is something to adopt, but let me check that.

voldemortX · 2020-10-28T03:15:48Z

One more thought that needs checking: affine(), rotate() and perspective() all have no inter-channel operations, so they could all support a n-tuple for fill color (other than a single float and int value). I'm not so sure since I find the docstring for F.affine and F_pil.affine states only int for fillcolor.

If this is correct, I'll try and support n-tuple as well.

Btw, do I need to change base class to test for float images, e.g. a new function _create_float_data() or a new metric for comparing float outputs.

vfdev-5 · 2020-10-28T09:00:28Z

@voldemortX i was also thinking about supporting tuple of values. Technically, it is not that complicated, however for instance, let's accomplish this feature with single int or float and work on tuple in another PR.

Btw, do I need to change base class to test for float images, e.g. a new function _create_float_data() or a new metric for comparing float outputs.

Currently, we are doing simply cast to float like here :

vision/test/test_functional_tensor.py

Line 590 in 1fe1e11

tensor = tensor.to(dtype=dt)

so no need to a new function.

voldemortX · 2020-10-28T09:13:15Z

@voldemortX i was also thinking about supporting tuple of values. Technically, it is not that complicated, however for instance, let's accomplish this feature with single int or float and work on tuple in another PR.

Btw, do I need to change base class to test for float images, e.g. a new function _create_float_data() or a new metric for comparing float outputs.

Currently, we are doing simply cast to float like here :

vision/test/test_functional_tensor.py

Line 590 in 1fe1e11

tensor = tensor.to(dtype=dt)

so no need to a new function.

Great, I'll commit for single value soon.

facebook-github-bot · 2020-10-31T07:32:15Z

Hi @voldemortX!

Thank you for your pull request and welcome to our community. We require contributors to sign our Contributor License Agreement, and we don't seem to have you on file.

In order for us to review and merge your code, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

If you have received this in error or have any questions, please contact us at [email protected]. Thanks!

vfdev-5 · 2020-11-02T15:38:43Z

@voldemortX I checked a bit more and yes img * mask + (1 - mask) * fillcolor is a better solution for linear interpolation. However, it wont match PIL's behaviour, but I think this is not a problem.

vfdev-5

Thanks for the update @voldemortX !
I left few comments to update it a little more.

torchvision/transforms/transforms.py

vfdev-5 · 2020-11-03T15:36:11Z

@voldemortX let's try to keep it simple. In addition if we would like to support list of int, float, let's do it now like that:

from typing import List
import torch

class A(torch.nn.Module):   
    def __init__(self, fill=0):
        super().__init__()
        # fill can be int, float or list
        # for torchscript: single value should be in the list: [value, ]
        self.fill = fill
        
    def forward(self, x):
        
        fill = self.fill
        if isinstance(x, torch.Tensor) and isinstance(fill, (int, float)):
            fill = [float(fill), ]
        return func(x, fill)
    
def func(x: torch.Tensor, fill: List[float]):    
    if not isinstance(x, torch.Tensor):
        print("Call Pillow with fill: {}".format(fill))
        return
    # dummy tensor func implementation : 
    y = x.clone()
    mask = x < 0.5 * x.max()    
    y[mask] = torch.tensor(fill, dtype=x.dtype)
    return y

For PIL, it should not change anything. Type hint for fill in F.rotate/affine will be only for torchscript. For tensor input we will cast in all functions fill = [float(fill), ] such that torchscript should pass.

Btw, I'll trying to land the following PR (#2952) with uniformized arg names, so we will have to take that into account.

voldemortX · 2020-11-04T12:11:40Z

@voldemortX let's try to keep it simple. In addition if we would like to support list of int, float, let's do it now like that:

from typing import List
import torch

class A(torch.nn.Module):   
    def __init__(self, fill=0):
        super().__init__()
        # fill can be int, float or list
        # for torchscript: single value should be in the list: [value, ]
        self.fill = fill
        
    def forward(self, x):
        
        fill = self.fill
        if isinstance(x, torch.Tensor) and isinstance(fill, (int, float)):
            fill = [float(fill), ]
        return func(x, fill)
    
def func(x: torch.Tensor, fill: List[float]):    
    if not isinstance(x, torch.Tensor):
        print("Call Pillow with fill: {}".format(fill))
        return
    # dummy tensor func implementation : 
    y = x.clone()
    mask = x < 0.5 * x.max()    
    y[mask] = torch.tensor(fill, dtype=x.dtype)
    return y

For PIL, it should not change anything. Type hint for fill in F.rotate/affine will be only for torchscript. For tensor input we will cast in all functions fill = [float(fill), ] such that torchscript should pass.

Btw, I'll trying to land the following PR (#2952) with uniformized arg names, so we will have to take that into account.

Should we just "revert" to the previous commit or implement for List in this PR also?

vfdev-5 · 2020-11-04T12:13:19Z

@voldemortX I'd say implement for List in this PR.

voldemortX · 2020-11-04T12:16:26Z

Okay! I'll do it soon with some new unit tests.

voldemortX · 2020-11-04T14:31:54Z

@vfdev-5 The documentation is n-tuple, should we do Tuple or List?

vfdev-5 · 2020-11-04T14:33:15Z

@voldemortX we wont be able to support single value with Tuple I think. Let's make it as List.

voldemortX · 2020-11-04T14:35:07Z

@voldemortX we wont be able to support single value with Tuple I think. Let's make it as List.

Should the doc be changed then?

vfdev-5

@voldemortX thanks a lot for the update !
I left a nit comment. Otherwise looks good!

torchvision/transforms/functional_tensor.py

vfdev-5

@voldemortX I have few other comments on testing and docstrings.
As we modify torchvision/transforms/transforms.py here, we also have to add few tests to torchvision/test/test_transforms_tensor.py

Thanks for working on this PR and appologies about how much time it take to make the merge.

vfdev-5 · 2020-11-30T13:18:25Z

torchvision/transforms/functional_pil.py

+                   "bands of the image ({} != {})")
+            raise ValueError(msg.format(len(fill), num_bands))
+        else:
+            fill = tuple(fill)


@voldemortX seems like I missed this modification in the previous review. Why do we need to modify the code here ? It's a PIL side and I think it can be kept as is ?

Don't quite remember changing that actually... I'll get it back to what it was.

Oh I remember now. I need the tuple(fill) conversion since now the input is formatted as List, also that means I can't do tests with tuple fill inputs. @vfdev-5

I see. You are right ! Pillow accepts only tuples and we can give a list too. Maybe, we can make the check more straightforward :

if isinstance(fill, (list, tuple)): if len(fill) != num_bands: msg = ("The number of elements in 'fill' does not match the number of " "bands of the image ({} != {})") raise ValueError(msg.format(len(fill), num_bands)) fill = tuple(fill)

Sure! I'll do that together in the next commit.

vfdev-5 · 2020-11-30T13:27:13Z

torchvision/transforms/functional_tensor.py

+        mask = mask.expand_as(img)
+        fill_img = torch.tensor(fill, dtype=img.dtype, device=img.device).view(1, len(fill), 1, 1).expand_as(img)
+        if mode == 'nearest':
+            img[mask < 1e-3] = fill_img[mask < 1e-3]  # Leave some room for error


I have a question here. Appologies if we've already discussed that. Technically, we have the mask as img dtype and we recreate boolean tensor. And we have to compare to 1 or 0.5, right ?

img[mask < 0.5] # or img[mask < 1.0]

why the threshold is 1-e3 here ?

In nearest interpolation, the mask should be all 0/1. So anything between (0-1) should be right? I use this to compensate for possible rounding errors from CUDA or something.

In nearest the result will be already rounded and we only have to get back bool mask values. I think we can either do img[mask < 0.5] or mask = mask.bool(). Let's do:

mask = mask < 0.5 img[mask] = fill_img[mask]

also without computing twice the binary mask

Yes, that seems better and faster.

vfdev-5 · 2020-11-30T13:32:56Z

test/test_functional_tensor.py

+            (33, (5, -4), 1.0, [0.0, 0.0], [0, 0, 0]),
+            (45, [-5, 4], 1.2, [0.0, 0.0], [1, 2, 3]),
+            (33, (-4, -8), 2.0, [0.0, 0.0], [255, 255, 255]),
+            (85, (10, -10), 0.7, [0.0, 0.0], None),


Let's add a test here with a single int and float as fill value as [a_int, ] and (b_float, ).

Right! I'll add the tests and docstring changes together in the next commit.

vfdev-5 · 2020-11-30T13:33:33Z

test/test_functional_tensor.py

-                                0.03,
-                                msg="{}: {}\n{} vs \n{}".format(
-                                    (img_size, r, dt, a, e, c),
+                        for f in [None, [0, 0, 0], [1, 2, 3], [255, 255, 255]]:


Same here, let's add single int and float values (as [a_int, ] and (b_float, ))

vfdev-5 · 2020-11-30T13:46:57Z

torchvision/transforms/functional.py

@@ -573,10 +573,9 @@ def perspective(
            :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.BILINEAR``.
            If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
-        fill (n-tuple or int or float): Pixel fill value for area outside the rotated
+        fill (sequence or int or float, optional): Pixel fill value for area outside the rotated


Let's make the docstring more explicit about how it works for PIL, tensor and torchscript:

fill (sequence or int or float, optional): Pixel fill value for the area outside the rotated image. If int or float, the value is used for all bands respectively. This option is supported for PIL image and Tensor inputs. In torchscript mode single int/float value is not supported, please use a tuple or list of length 1: ``[value, ]``. If input is PIL Image, the options is only available for ``Pillow>=5.0.0``.

same for other docstrings.

@vfdev-5 I did some sorting. It seems the current version means affine functions support:
int/float/list/None for Tensor.
int/float/list/tuple/None for PIL.
list/None for torchscript.
Not sure what I should write in the docs. (Tensor and PIL seems still have this difference on tuple)

I thought PIL and Tensor could support the same types. Where is actually the problem with tuple if input is Tensor ?

Not very sure, let me test for it a bit. I wrote the code with only list in mind.

You are right, torch.tensor() can also convert tuples. There is no functional mismatch between tensor and PIL.

torchvision/transforms/functional_tensor.py

voldemortX · 2020-11-30T15:59:03Z

@vfdev-5 This version should cover things, I rigged some new unit tests. But, why the macos tests always fail...

vfdev-5 · 2020-11-30T20:07:05Z

@voldemortX macosx tests failure is unrelated. However, I agree it is annoying to see the CI failing. Probably, it will be fixed in the coming days.
As for tests, as we modify torchvision/transforms/transforms.py in this PR, could you please add few tests to torchvision/test/test_transforms_tensor.py too to test modifed classes from transforms.py.

(85, (10, -10), 0.7, [0.0, 0.0], [1, ], 1)

the way you passing tensor and pil config is OK, but I'd pass either [1, ] or 1 and in the code adapt for PIL or Tensor respectively the value...

…anges

voldemortX · 2020-12-01T03:57:23Z

Thanks for the review! I added the tests and really found some bugs and rectified them, specifically more compact support for tuples/lists and a check for num_channels like in PIL the old codes check for num_bands. Only for tensors, [x, ] also works same as single number due to it can be broadcasted to match the image.

I also find that in transforms.py we casted everything to list[float], so torchscript should also support every input types here. So I changed the docs a bit there.

p.s. To cast test fill values for tensor and PIL images respectively in the code and not in the configs is possible for test_functional_tensor.py, do you mean I should do a conversion like f_pil = int(f[0]) if len(f) == 1 else f ?

vfdev-5 · 2020-12-01T07:39:43Z

@voldemortX thanks for the update !

really found some bugs and rectified them, specifically more compact support for tuples/lists and a check for num_channels like in PIL the old codes check for num_bands.

These bugs are in transforms.py only ?

do you mean I should do a conversion like f_pil = int(f[0]) if len(f) == 1 else f ?

Yes, something like that can work. Or like that:

test_config = (85, (10, -10), 0.7, [0.0, 0.0], 1),

f_t = [f, ] if isinstance(f, (int, float)) else f

voldemortX · 2020-12-01T10:05:57Z

These bugs are in transforms.py only ?

I directly ran (85, (10, -10), 0.7, [0.0, 0.0], 1) and found some bugs. All changes in the last commit excluding the tests are addressing bugs. One outside transforms.py is the num_channels check. But they should be alright now.

EDIT: I'd like to go with f_pil in the tests, since tensor fills have more freedom. e.g. PIL only accept int/tuple(int)/list[int]... for int images. Seems that is the usual behavior of PIL.

test/test_functional_tensor.py

vfdev-5 · 2020-12-01T11:52:53Z

@voldemortX thanks a lot for the update ! Looks good now.
Let me do another pass a bit later and probably it will be good to merge...

vfdev-5

@voldemortX few more nits and I think we are good for this PR. Thanks for working on this PR !

torchvision/transforms/functional_tensor.py

torchvision/transforms/transforms.py

vfdev-5

Thanks a lot @voldemortX !

* Fill color support for tensor affine transforms * PEP fix * Docstring changes and float support * Docstring update for transforms and float type cast * Cast only for Tensor * Temporary patch for lack of Union type support, plus an extra unit test * More plausible bilinear filling for tensors * Keep things simple & New docstrings * Fix lint and other issues after merge * make it in one line * Docstring and some code modifications * More tests and corresponding changes for transoforms and docstring changes * Simplify test configs * Update test_functional_tensor.py * Update test_functional_tensor.py * Move assertions Co-authored-by: vfdev <[email protected]>

Summary: * Fill color support for tensor affine transforms * PEP fix * Docstring changes and float support * Docstring update for transforms and float type cast * Cast only for Tensor * Temporary patch for lack of Union type support, plus an extra unit test * More plausible bilinear filling for tensors * Keep things simple & New docstrings * Fix lint and other issues after merge * make it in one line * Docstring and some code modifications * More tests and corresponding changes for transoforms and docstring changes * Simplify test configs * Update test_functional_tensor.py * Update test_functional_tensor.py * Move assertions Reviewed By: datumbox Differential Revision: D25396712 fbshipit-source-id: 7eb32024c91b67ffa154a481aa592c6e57b3c480 Co-authored-by: vfdev <[email protected]>

voldemortX and others added 4 commits October 27, 2020 20:42

Fill color support for tensor affine transforms

696c15a

Merge branch 'master' into issue2887

229c140

PEP fix

adae0f6

Merge branch 'issue2887' of github.com:voldemortX/vision into issue2887

b2721e8

vfdev-5 reviewed Oct 27, 2020

View reviewed changes

voldemortX added 4 commits October 28, 2020 20:21

Docstring changes and float support

1c4e48a

Docstring update for transforms and float type cast

62abb37

Cast only for Tensor

a585dbd

Temporary patch for lack of Union type support, plus an extra unit test

d616210

facebook-github-bot added the cla signed label Oct 31, 2020

vfdev-5 reviewed Nov 2, 2020

View reviewed changes

torchvision/transforms/transforms.py Outdated Show resolved Hide resolved

torchvision/transforms/transforms.py Outdated Show resolved Hide resolved

More plausible bilinear filling for tensors

417f6ea

vfdev-5 reviewed Nov 30, 2020

View reviewed changes

torchvision/transforms/functional_tensor.py Outdated Show resolved Hide resolved

vfdev-5 marked this pull request as ready for review November 30, 2020 11:16

voldemortX and others added 2 commits November 30, 2020 19:45

make it in one line

731a5a9

Merge branch 'master' into issue2887

4389f80

vfdev-5 reviewed Nov 30, 2020

View reviewed changes

voldemortX added 2 commits November 30, 2020 23:29

Docstring and some code modifications

2ea1003

Merge branch 'issue2887' of github.com:voldemortX/vision into issue2887

4c59964

More tests and corresponding changes for transoforms and docstring ch…

9e7cb7a

…anges

Simplify test configs

16e9b97

Update test_functional_tensor.py

96c70bc

vfdev-5 reviewed Dec 1, 2020

View reviewed changes

test/test_functional_tensor.py Outdated Show resolved Hide resolved

Update test_functional_tensor.py

9d9fd08

vfdev-5 reviewed Dec 2, 2020

View reviewed changes

torchvision/transforms/functional_tensor.py Outdated Show resolved Hide resolved

torchvision/transforms/transforms.py Outdated Show resolved Hide resolved

vfdev-5 and others added 2 commits December 2, 2020 09:08

Merge branch 'master' into issue2887

87560cb

Move assertions

bc7e9fe

vfdev-5 approved these changes Dec 2, 2020

View reviewed changes

vfdev-5 merged commit 21deb4d into pytorch:master Dec 2, 2020

voldemortX deleted the issue2887 branch December 2, 2020 11:01

vfdev-5 mentioned this pull request Jan 27, 2021

transforms.RandomRotation with Tensors leads to an error #3309

Closed

vfdev-5 mentioned this pull request Sep 27, 2021

Harmonize functional transforms arguments #2479

Closed

vfdev-5 mentioned this pull request Feb 15, 2023

make fill defaultdict an implementation detail #7258

Merged

vfdev-5 mentioned this pull request Nov 8, 2023

remove fill blending for bilinear affine #8098

Open

Fill color support for tensor affine transforms #2904

Fill color support for tensor affine transforms #2904

Conversation

voldemortX commented Oct 27, 2020 • edited Loading

vfdev-5 left a comment

Choose a reason for hiding this comment

vfdev-5 commented Oct 27, 2020

voldemortX commented Oct 27, 2020

vfdev-5 commented Oct 27, 2020

voldemortX commented Oct 27, 2020 • edited Loading

codecov bot commented Oct 27, 2020 • edited Loading

Codecov Report

vfdev-5 commented Oct 27, 2020

voldemortX commented Oct 28, 2020

vfdev-5 commented Oct 28, 2020

voldemortX commented Oct 28, 2020

facebook-github-bot commented Oct 31, 2020

vfdev-5 commented Nov 2, 2020 • edited Loading

vfdev-5 left a comment

Choose a reason for hiding this comment

vfdev-5 commented Nov 3, 2020

voldemortX commented Nov 4, 2020 • edited Loading

vfdev-5 commented Nov 4, 2020

voldemortX commented Nov 4, 2020

voldemortX commented Nov 4, 2020

vfdev-5 commented Nov 4, 2020

voldemortX commented Nov 4, 2020

vfdev-5 left a comment

Choose a reason for hiding this comment

vfdev-5 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

voldemortX Nov 30, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vfdev-5 Nov 30, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

voldemortX Nov 30, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

voldemortX commented Nov 30, 2020

vfdev-5 commented Nov 30, 2020

voldemortX commented Dec 1, 2020 • edited Loading

vfdev-5 commented Dec 1, 2020

voldemortX commented Dec 1, 2020 • edited Loading

vfdev-5 commented Dec 1, 2020

vfdev-5 left a comment • edited Loading

Choose a reason for hiding this comment

vfdev-5 left a comment

Choose a reason for hiding this comment

voldemortX commented Oct 27, 2020 •

edited

Loading

voldemortX commented Oct 27, 2020 •

edited

Loading

codecov bot commented Oct 27, 2020 •

edited

Loading

vfdev-5 commented Nov 2, 2020 •

edited

Loading

voldemortX commented Nov 4, 2020 •

edited

Loading

voldemortX Nov 30, 2020 •

edited

Loading

vfdev-5 Nov 30, 2020 •

edited

Loading

voldemortX Nov 30, 2020 •

edited

Loading

voldemortX commented Dec 1, 2020 •

edited

Loading

voldemortX commented Dec 1, 2020 •

edited

Loading

vfdev-5 left a comment •

edited

Loading