Added nearest-exact interpolation mode by vfdev-5 · Pull Request #64501 · pytorch/pytorch

vfdev-5 · 2021-09-03T22:53:15Z

Added "nearest-exact" interpolation mode to fix the issues: #34808 and #62237.

Description:

As we can not fix "nearest" mode without large impact on already trained model it was suggested to introduce new mode instead of fixing exising "nearest" mode.

New mode "nearest-exact" performs index computation for nearest interpolation to match scikit-image, pillow, TF2 and while "nearest" mode still match opencv INTER_NEAREST, which appears to be buggy, see https://ppwwyyxx.com/blog/2021/Where-are-Pixels/#Libraries.

"nearest":

input_index_f32 = output_index * scale
input_index = floor(input_index_f32)

"nearest-exact"

input_index_f32 = (output_index + 0.5) * scale - 0.5
input_index = round(input_index_f32)

Comparisions with other libs: https://gist.github.com/vfdev-5/a5bd5b1477b1c82a87a0f9e25c727664

PyTorch version	1.9.0 "nearest"	this PR "nearest"	this PR "nearest-exact"
Resize option:
OpenCV INTER_NEAREST result mismatches	0	0	10
OpenCV INTER_NEAREST_EXACT result mismatches	9	9	9
Scikit-Image result mismatches	10	10	0
Pillow result mismatches	10	10	7
TensorFlow result mismatches	10	10	0
Rescale option:
size mismatches (#62396)	10	10	10
OpenCV INTER_NEAREST result mismatches	3	3	5
OpenCV INTER_NEAREST_EXACT result mismatches	3	3	4
Scikit-Image result mismatches	4	4	0
Scipy result mismatches	4	4	0
TensorFlow: no such option	-	-

Versions:

skimage: 0.19.0.dev0
opencv: 4.5.4-dev
scipy: 1.7.2
Pillow: 8.4.0
TensorFlow: 2.7.0

Implementations in other libs:

Additionally:

Updated upsampling tests

cc @ezyang @gchanan @albanD @mruberry @jbschlosser @walterddr @fmassa @heitorschueroff @ppwwyyxx

…v-5/fix-nearest-upsampling

- updated index computation formula to match scikit-image/scipy Fixes pytorch#34808, pytorch#62237

facebook-github-bot · 2021-09-03T22:53:21Z

🔗 Helpful links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/64501
📄 Preview docs built from this PR
📄 Preview C++ docs built from this PR
🔧 Opt-in to CIFlow to control what jobs run on your PRs

💊 CI failures summary and remediations

As of commit a612bb1 (more details on the Dr. CI page):

💚 💚 Looks good so far! There are no failures yet. 💚 💚

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

ppwwyyxx · 2021-09-03T23:02:40Z

FYI I just noticed that opencv already fixed the issue with a new interpolation mode: opencv/opencv#18053

vfdev-5 · 2021-09-07T14:27:42Z

opencv already fixed the issue with a new interpolation mode: opencv/opencv#18053

@ppwwyyxx I have an impression that it gives other results and not always matches PIL or Scikit-Image (which not always match neither).
For example

import numpy as np
from PIL import Image
import cv2
from skimage.transform import resize

size = 10
a_in = np.arange(0, 3 * 32 * 32, dtype="uint8").reshape(32, 32, 3)

pil_in = Image.fromarray(a_in)
pil_out = pil_in.resize((size, size), resample=Image.NEAREST)
a_pil_out = np.array(pil_out)

a_cv2_out = cv2.resize(a_in, dsize=(size, size), interpolation=cv2.INTER_NEAREST_EXACT)

a_skimg_out = resize(a_in, (size, size), order=0, preserve_range=True, anti_aliasing=False).astype("uint8")

print(np.allclose(a_cv2_out, a_pil_out), np.mean(np.abs(a_cv2_out - a_pil_out)))
# > (False, 38.74)

print(np.allclose(a_cv2_out, a_skimg_out), np.mean(np.abs(a_cv2_out - a_skimg_out)))
# > (False, 25.3)

fmassa

Changes LGTM, thanks!

But due to BC-breakages on the numerics of the function, I'll defer to @jbschlosser to decide how to handle this.

ppwwyyxx · 2021-09-08T22:10:42Z

@vfdev-5 From what I can tell the difference is only due to half-rounding: when a sample fall into the middle point between two pixels, it's ambiguous which pixel is its nearest neighbor, and then different implementations make different choices.

The below script verify that for all (input_size, output_size) in (1, 100), all the differences between cv2(NEAREST_EXACT), PIL and skimage are due to the above reason. In other words, these three libraries are in full agreement except for this unavoidable ambiguity of nearest neighbor resize. Please also test this against your PR if possible.

import numpy as np
from PIL import Image
import cv2
from skimage.transform import resize
np.set_printoptions(threshold=9888888)

for size in range(1, 100):
    for isize in range(1, 100):
        a_in = (np.random.rand(isize, isize) * 255).astype("uint8")

        pil_in = Image.fromarray(a_in)
        pil_out = pil_in.resize((size, size), resample=Image.NEAREST)
        a_pil_out = np.array(pil_out)

        a_cv2_out = cv2.resize(a_in, dsize=(size, size), interpolation=cv2.INTER_NEAREST_EXACT)

        a_skimg_out = resize(a_in, (size, size), order=0, mode='edge',
                preserve_range=True, anti_aliasing=False).astype("uint8")

        def is_int(x):
            return np.allclose(np.round(x), x)

        def find_diff(a, b):
            diff = a - b
            if np.abs(diff).max() < 1e-6:
                return
            for y, x in zip(*np.nonzero(diff)):
                oy = (1 / size) * (y + 0.5) / (1 / isize)
                ox = (1 / size) * (x + 0.5) / (1 / isize)
                assert is_int(oy) | is_int(ox), (oy, ox)

        find_diff(a_cv2_out, a_pil_out)
        find_diff(a_cv2_out, a_skimg_out)
        find_diff(a_pil_out, a_skimg_out)

vfdev-5 · 2021-09-08T23:34:06Z

Thanks @ppwwyyxx for pointing out that. Yes, you are right, I see what you mean. I think more precisely the point is about int(float_index) where float index has no floating point and producing either i - 1 or i.

isize = 32
size = 10

scale = 1.0 * isize / size
for o in range(size):
    i_f32 = (o + 0.5) * scale
    i = int(i_f32)
    print(o, " <- ", i, i_f32)

> 
0  <-  1 1.6
1  <-  4 4.800000000000001
2  <-  8 8.0
3  <-  11 11.200000000000001
4  <-  14 14.4
5  <-  17 17.6
6  <-  20 20.8
7  <-  24 24.0
8  <-  27 27.200000000000003
9  <-  30 30.400000000000002

a_in = np.arange(0, isize, dtype="float32")[:, None]
pil_in = Image.fromarray(a_in)
pil_out = pil_in.resize((1, size), resample=Image.NEAREST)
a_pil_out = np.array(pil_out)
a_cv2_out = cv2.resize(a_in, dsize=(1, size), interpolation=cv2.INTER_NEAREST_EXACT)
a_skimg_out = resize(a_in, (size, 1), order=0, preserve_range=True, anti_aliasing=False).astype(a_in.dtype)

print("PIL", "SKI", "CV")
print(a_pil_out[:, 0])
print(a_skimg_out[:, 0])
print(a_cv2_out[:, 0])
>
PIL SKI CV
[ 1.  4.  8. 11. 14. 17. 20. 23. 27. 30.]
[ 1.  4.  8. 11. 14. 17. 20. 24. 27. 30.]
[ 1.  4.  7. 11. 14. 17. 20. 23. 27. 30.]

jbschlosser

But due to BC-breakages on the numerics of the function, I'll defer to @jbschlosser to decide how to handle this.

I'm wary of making this change all at once, since it would certainly require model retraining to avoid performance metric regressions from preprocessing / trained model mismatch.

opencv already fixed the issue with a new interpolation mode: opencv/opencv#18053

I think the way to go here is to follow suit and add a new interpolation mode as well to maintain BC. It's unfortunate but I don't see a way around it while still avoiding lots of painful breakage.

vfdev-5 · 2021-09-09T08:17:26Z

Thanks for your comment @jbschlosser !

I think the way to go here is to follow suit and add a new interpolation mode as well to maintain BC. It's unfortunate but I don't see a way around it while still avoiding lots of painful breakage.

Seems reasonable. For example, new mode can be called "nearest-exact" or something similar. Maybe, we can keep both "nearest" and "nearest-exact" in v1.X and remove "nearest-exact", keep "nearest" and replace buggy code for v2.X (as TF did according to https://ppwwyyxx.com/blog/2021/Where-are-Pixels/#Libraries)

Any better suggestions for the mode name, e.g "nearest-exact" ?

…v-5/fix-nearest-upsampling

- upsample_nearest_exact1d

and tests

ppwwyyxx · 2021-09-10T21:24:13Z

I'm not so sure why opencv chooses to call it "Bit-Exact". From opencv/opencv#18053 (comment) it appears it means "it produces bitwise identical output regardless of platform". If that's what the name means that doesn't sound like what we're looking for.
From my point of view the old "nearest" is just wrong (in both opencv and pytorch) and the new one would be "nearest-correct" or something.

pytorch-probot · 2021-11-08T13:54:30Z

CI Flow Status

⚛️ CI Flow

Ruleset - Version: v1
Ruleset - File: https://github.com/Quansight/pytorch/blob/0de5467ea8f7d46257e501e3da001e3feabc64b3/.github/generated-ciflow-ruleset.json
PR ciflow labels: ciflow/default

Workflows	Labels (bold enabled)	Status
Triggered Workflows
linux-bionic-py3.6-clang9	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/noarch`, `ciflow/xla`	✅ triggered
linux-vulkan-bionic-py3.6-clang9	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/vulkan`	✅ triggered
linux-xenial-cuda11.3-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3-clang5-mobile-build	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3-clang5-mobile-custom-build-dynamic	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3-clang5-mobile-custom-build-static	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3.6-clang7-asan	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/sanitizers`	✅ triggered
linux-xenial-py3.6-clang7-onnx	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/onnx`	✅ triggered
linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3.6-gcc7	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3.6-gcc7-bazel-test	`ciflow/all`, `ciflow/bazel`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
pytorch-linux-xenial-py3-clang5-android-ndk-r19c-gradle-custom-build-single	`ciflow/all`, `ciflow/android`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
pytorch-linux-xenial-py3-clang5-android-ndk-r19c-gradle-custom-build-single-full-jit	`ciflow/all`, `ciflow/android`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
win-vs2019-cpu-py3	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/win`	✅ triggered
win-vs2019-cuda11.3-py3	`ciflow/all`, `ciflow/cuda`, `ciflow/default`, `ciflow/win`	✅ triggered
Skipped Workflows
caffe2-linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/linux`	🚫 skipped
docker-builds	`ciflow/all`	🚫 skipped
ios-12-5-1-arm64	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-coreml	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-custom-ops	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-full-jit	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-metal	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64-coreml	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64-full-jit	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
libtorch-linux-xenial-cuda10.2-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`	🚫 skipped
libtorch-linux-xenial-cuda11.3-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`	🚫 skipped
linux-bionic-cuda10.2-py3.9-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/slow`	🚫 skipped
linux-xenial-py3-clang5-mobile-code-analysis	`ciflow/all`, `ciflow/linux`, `ciflow/mobile`	🚫 skipped
parallelnative-linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/linux`	🚫 skipped
periodic-libtorch-linux-xenial-cuda11.1-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`, `ciflow/scheduled`	🚫 skipped
periodic-linux-xenial-cuda10.2-py3-gcc7-slow-gradcheck	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/scheduled`, `ciflow/slow`, `ciflow/slow-gradcheck`	🚫 skipped
periodic-linux-xenial-cuda11.1-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/scheduled`	🚫 skipped
periodic-win-vs2019-cuda11.1-py3	`ciflow/all`, `ciflow/cuda`, `ciflow/scheduled`, `ciflow/win`	🚫 skipped

You can add a comment to the PR and tag @pytorchbot with the following commands:

# ciflow rerun, "ciflow/default" will always be added automatically
@pytorchbot ciflow rerun

# ciflow rerun with additional labels "-l <ciflow/label_name>", which is equivalent to adding these labels manually and trigger the rerun
@pytorchbot ciflow rerun -l ciflow/scheduled -l ciflow/slow

For more information, please take a look at the CI Flow Wiki.

vfdev-5 · 2021-11-09T12:34:30Z

@jbschlosser I updated the PR according to your previous recommendations: a) separate private ops doing the nearest exact interpolation and b) added work around for JIT FC. Could you please review the PR. Thanks!

jbschlosser

Looking pretty good, just a minor script-gating thing :)

torch/nn/functional.py

pytorch-probot · 2021-11-10T10:33:49Z

CI Flow Status

⚛️ CI Flow

Ruleset - Version: v1
Ruleset - File: https://github.com/Quansight/pytorch/blob/669817e71c67b8809c93271618edfacce35da815/.github/generated-ciflow-ruleset.json
PR ciflow labels: ciflow/default

Workflows	Labels (bold enabled)	Status
Triggered Workflows
linux-bionic-py3.6-clang9	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/noarch`, `ciflow/xla`	✅ triggered
linux-vulkan-bionic-py3.6-clang9	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/vulkan`	✅ triggered
linux-xenial-cuda11.3-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3-clang5-mobile-build	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3-clang5-mobile-custom-build-dynamic	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3-clang5-mobile-custom-build-static	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3.6-clang7-asan	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/sanitizers`	✅ triggered
linux-xenial-py3.6-clang7-onnx	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/onnx`	✅ triggered
linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3.6-gcc7	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3.6-gcc7-bazel-test	`ciflow/all`, `ciflow/bazel`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
pytorch-linux-xenial-py3-clang5-android-ndk-r19c-gradle-custom-build-single	`ciflow/all`, `ciflow/android`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
pytorch-linux-xenial-py3-clang5-android-ndk-r19c-gradle-custom-build-single-full-jit	`ciflow/all`, `ciflow/android`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
win-vs2019-cpu-py3	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/win`	✅ triggered
win-vs2019-cuda11.3-py3	`ciflow/all`, `ciflow/cuda`, `ciflow/default`, `ciflow/win`	✅ triggered
Skipped Workflows
caffe2-linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/linux`	🚫 skipped
docker-builds	`ciflow/all`	🚫 skipped
ios-12-5-1-arm64	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-coreml	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-custom-ops	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-full-jit	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-metal	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64-coreml	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64-full-jit	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
libtorch-linux-xenial-cuda10.2-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`	🚫 skipped
libtorch-linux-xenial-cuda11.3-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`	🚫 skipped
linux-bionic-cuda10.2-py3.9-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/slow`	🚫 skipped
linux-xenial-py3-clang5-mobile-code-analysis	`ciflow/all`, `ciflow/linux`, `ciflow/mobile`	🚫 skipped
parallelnative-linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/linux`	🚫 skipped
periodic-libtorch-linux-xenial-cuda11.1-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`, `ciflow/scheduled`	🚫 skipped
periodic-linux-xenial-cuda10.2-py3-gcc7-slow-gradcheck	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/scheduled`, `ciflow/slow`, `ciflow/slow-gradcheck`	🚫 skipped
periodic-linux-xenial-cuda11.1-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/scheduled`	🚫 skipped
periodic-win-vs2019-cuda11.1-py3	`ciflow/all`, `ciflow/cuda`, `ciflow/scheduled`, `ciflow/win`	🚫 skipped

You can add a comment to the PR and tag @pytorchbot with the following commands:

# ciflow rerun, "ciflow/default" will always be added automatically
@pytorchbot ciflow rerun

# ciflow rerun with additional labels "-l <ciflow/label_name>", which is equivalent to adding these labels manually and trigger the rerun
@pytorchbot ciflow rerun -l ciflow/scheduled -l ciflow/slow

For more information, please take a look at the CI Flow Wiki.

…v-5/fix-nearest-upsampling

pytorch-probot · 2021-11-10T14:04:19Z

CI Flow Status

⚛️ CI Flow

Ruleset - Version: v1
Ruleset - File: https://github.com/Quansight/pytorch/blob/a612bb1bb3729d8e42f7cce94cc1d0bcc9cae148/.github/generated-ciflow-ruleset.json
PR ciflow labels: ciflow/default

Workflows	Labels (bold enabled)	Status
Triggered Workflows
linux-bionic-py3.6-clang9	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/noarch`, `ciflow/xla`	✅ triggered
linux-vulkan-bionic-py3.6-clang9	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/vulkan`	✅ triggered
linux-xenial-cuda11.3-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3-clang5-mobile-build	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3-clang5-mobile-custom-build-dynamic	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3-clang5-mobile-custom-build-static	`ciflow/all`, `ciflow/default`, `ciflow/linux`, `ciflow/mobile`	✅ triggered
linux-xenial-py3.6-clang7-asan	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/sanitizers`	✅ triggered
linux-xenial-py3.6-clang7-onnx	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`, `ciflow/onnx`	✅ triggered
linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3.6-gcc7	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
linux-xenial-py3.6-gcc7-bazel-test	`ciflow/all`, `ciflow/bazel`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
pytorch-linux-xenial-py3-clang5-android-ndk-r19c-gradle-custom-build-single	`ciflow/all`, `ciflow/android`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
pytorch-linux-xenial-py3-clang5-android-ndk-r19c-gradle-custom-build-single-full-jit	`ciflow/all`, `ciflow/android`, `ciflow/cpu`, `ciflow/default`, `ciflow/linux`	✅ triggered
win-vs2019-cpu-py3	`ciflow/all`, `ciflow/cpu`, `ciflow/default`, `ciflow/win`	✅ triggered
win-vs2019-cuda11.3-py3	`ciflow/all`, `ciflow/cuda`, `ciflow/default`, `ciflow/win`	✅ triggered
Skipped Workflows
caffe2-linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/linux`	🚫 skipped
docker-builds	`ciflow/all`	🚫 skipped
ios-12-5-1-arm64	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-coreml	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-custom-ops	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-full-jit	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-arm64-metal	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64-coreml	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
ios-12-5-1-x86-64-full-jit	`ciflow/all`, `ciflow/ios`, `ciflow/macos`	🚫 skipped
libtorch-linux-xenial-cuda10.2-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`	🚫 skipped
libtorch-linux-xenial-cuda11.3-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`	🚫 skipped
linux-bionic-cuda10.2-py3.9-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/slow`	🚫 skipped
linux-xenial-py3-clang5-mobile-code-analysis	`ciflow/all`, `ciflow/linux`, `ciflow/mobile`	🚫 skipped
macos-10-15-py3-arm64	`ciflow/all`, `ciflow/macos`	🚫 skipped
macos-10-15-py3-lite-interpreter-x86-64	`ciflow/all`, `ciflow/macos`	🚫 skipped
macos-10-15-py3-x86-64	`ciflow/all`, `ciflow/macos`	🚫 skipped
parallelnative-linux-xenial-py3.6-gcc5.4	`ciflow/all`, `ciflow/cpu`, `ciflow/linux`	🚫 skipped
periodic-libtorch-linux-xenial-cuda11.1-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/libtorch`, `ciflow/linux`, `ciflow/scheduled`	🚫 skipped
periodic-linux-xenial-cuda10.2-py3-gcc7-slow-gradcheck	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/scheduled`, `ciflow/slow`, `ciflow/slow-gradcheck`	🚫 skipped
periodic-linux-xenial-cuda11.1-py3.6-gcc7	`ciflow/all`, `ciflow/cuda`, `ciflow/linux`, `ciflow/scheduled`	🚫 skipped
periodic-win-vs2019-cuda11.1-py3	`ciflow/all`, `ciflow/cuda`, `ciflow/scheduled`, `ciflow/win`	🚫 skipped

You can add a comment to the PR and tag @pytorchbot with the following commands:

# ciflow rerun, "ciflow/default" will always be added automatically
@pytorchbot ciflow rerun

# ciflow rerun with additional labels "-l <ciflow/label_name>", which is equivalent to adding these labels manually and trigger the rerun
@pytorchbot ciflow rerun -l ciflow/scheduled -l ciflow/slow

For more information, please take a look at the CI Flow Wiki.

jbschlosser

Looks good on my end!

facebook-github-bot · 2021-11-11T18:10:05Z

@jbschlosser has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

facebook-github-bot · 2021-11-15T22:30:20Z

@jbschlosser merged this pull request in 6adbe04.

@cpuhrsch

Description: Following pytorch#65142 (comment) adding missing nearest-exact mode and anti-alias flag - pytorch#65142 - pytorch#64501 cc @cpuhrsch

Summary: Description: Following #65142 (comment) adding missing nearest-exact mode and anti-alias flag to C++ frontend. - #65142 - #64501 - added tests in pytorch/test/cpp/api/functional.cpp Pull Request resolved: #69318 Reviewed By: davidberard98 Differential Revision: D33278995 Pulled By: jbschlosser fbshipit-source-id: fa87c0c78df6b398e4f9688cc02111eed187afa7

Summary: Description: - Removed JIT FC tweaks for interpolation options : nearest-exact and antialiasing They were added in - #64501 (Sept 04 2021) - #65142 (Sept 16 2021) cc jbschlosser Pull Request resolved: #71937 Reviewed By: mrshenli Differential Revision: D33845502 Pulled By: jbschlosser fbshipit-source-id: 8a94454fd643cd2aef21b06689f72a0f16620d30

Summary: Description: - Removed JIT FC tweaks for interpolation options : nearest-exact and antialiasing They were added in - #64501 (Sept 04 2021) - #65142 (Sept 16 2021) cc jbschlosser Pull Request resolved: #71937 Reviewed By: mrshenli Differential Revision: D33845502 Pulled By: jbschlosser fbshipit-source-id: 8a94454fd643cd2aef21b06689f72a0f16620d30 (cherry picked from commit b21173d)

Summary: Description: - Removed JIT FC tweaks for interpolation options : nearest-exact and antialiasing They were added in - pytorch/pytorch#64501 (Sept 04 2021) - pytorch/pytorch#65142 (Sept 16 2021) cc jbschlosser Pull Request resolved: pytorch/pytorch#71937 Reviewed By: mrshenli Differential Revision: D33845502 Pulled By: jbschlosser fbshipit-source-id: 8a94454fd643cd2aef21b06689f72a0f16620d30 (cherry picked from commit b21173d)

vfdev-5 added 3 commits September 3, 2021 04:32

WIP on fixing nearest interpolation

a88520b

Merge branch 'master' of https://github.com/pytorch/pytorch into vfde…

1b6bc8d

…v-5/fix-nearest-upsampling

Fixed issues with nearest interpolation

ea32801

- updated index computation formula to match scikit-image/scipy Fixes pytorch#34808, pytorch#62237

vfdev-5 requested review from albanD and jbschlosser as code owners September 3, 2021 22:53

facebook-github-bot added the cla signed label Sep 3, 2021

pytorchbot added the open source label Sep 3, 2021

Fixed mypy issue

d735232

albanD removed their request for review September 6, 2021 21:24

mrshenli added module: nn Related to torch.nn triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module labels Sep 7, 2021

fmassa approved these changes Sep 8, 2021

View reviewed changes

albanD added the module: bc-breaking Related to a BC-breaking change label Sep 8, 2021

jbschlosser requested changes Sep 9, 2021

View reviewed changes

vfdev-5 added 4 commits September 9, 2021 03:51

Merge branch 'master' of https://github.com/pytorch/pytorch into vfde…

282406e

…v-5/fix-nearest-upsampling

WIP adding new mode nearest-exact to keep BC

257cffa

- upsample_nearest_exact1d

WIP on adding upsample_nearest_exact2d

fc87aea

and tests

WIP added upsample_nearest3d and upsample_nearest_exact3d + test

e6c0ff7

vfdev-5 requested review from ezyang and soulitzer as code owners September 10, 2021 13:45

Updated aten/src/ATen/core/aten_interned_strings.h

ccbdde5

vfdev-5 changed the title ~~[BC-breaking] Fixes nearest interpolation~~ Added nearest-exact interpolation mode Sep 10, 2021

vfdev-5 requested a review from jbschlosser September 10, 2021 14:21

Fixed CI issue for scripting

0de5467

jbschlosser reviewed Nov 9, 2021

View reviewed changes

torch/nn/functional.py Show resolved Hide resolved

Fixed torch.jit.is_scripting() gating as suggested

669817e

Merge branch 'master' of https://github.com/pytorch/pytorch into vfde…

a612bb1

…v-5/fix-nearest-upsampling

jbschlosser approved these changes Nov 10, 2021

View reviewed changes

facebook-github-bot closed this in 6adbe04 Nov 15, 2021

facebook-github-bot added the Merged label Nov 15, 2021

vfdev-5 deleted the vfdev-5/fix-nearest-upsampling branch November 15, 2021 22:31

vfdev-5 mentioned this pull request Nov 29, 2021

Added antialias flag to interpolate (CPU only, bilinear) #65142

Closed

vfdev-5 mentioned this pull request Dec 2, 2021

[C++ API] Added missing nearest-exact mode and anti-alias flag #69318

Closed

This was referenced Jan 7, 2022

torch.nn.functional.interpolate nearest behaviour changed in PyTorch 1.9 and appears incorrect #62237

Closed

Using torch.nn.functional.interpolate with 'nearest' mode introduces misalignment #34808

Closed

vfdev-5 mentioned this pull request Jan 27, 2022

Removed JIT FC tweaks for interpolation options #71937

Closed

pmeier mentioned this pull request Sep 26, 2022

Allow "nearest-exact" interpolation mode? pytorch/vision#6645

Open

vfdev-5 mentioned this pull request Apr 11, 2023

Fixed cuda grads for nearest interpolation with scale factor #97770

Closed

Conversation

vfdev-5 commented Sep 3, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

facebook-github-bot commented Sep 3, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful links

💊 CI failures summary and remediations

Uh oh!

ppwwyyxx commented Sep 3, 2021

Uh oh!

vfdev-5 commented Sep 7, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fmassa left a comment

Choose a reason for hiding this comment

Uh oh!

ppwwyyxx commented Sep 8, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vfdev-5 commented Sep 8, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jbschlosser left a comment

Choose a reason for hiding this comment

Uh oh!

vfdev-5 commented Sep 9, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ppwwyyxx commented Sep 10, 2021

Uh oh!

pytorch-probot bot commented Nov 8, 2021

⚛️ CI Flow

Uh oh!

vfdev-5 commented Nov 9, 2021

Uh oh!

jbschlosser left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

pytorch-probot bot commented Nov 10, 2021

⚛️ CI Flow

Uh oh!

pytorch-probot bot commented Nov 10, 2021

⚛️ CI Flow

Uh oh!

jbschlosser left a comment

Choose a reason for hiding this comment

Uh oh!

facebook-github-bot commented Nov 11, 2021

Uh oh!

facebook-github-bot commented Nov 15, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

8 participants

vfdev-5 commented Sep 3, 2021 •

edited

Loading

facebook-github-bot commented Sep 3, 2021 •

edited

Loading

vfdev-5 commented Sep 7, 2021 •

edited

Loading

ppwwyyxx commented Sep 8, 2021 •

edited

Loading

vfdev-5 commented Sep 8, 2021 •

edited

Loading

vfdev-5 commented Sep 9, 2021 •

edited

Loading