`SanitizeBoundingBox` based on minimum area #7735

antoinebrl · 2023-07-11T19:08:49Z

During transformations, bounding boxes can be filtered based on minimum size and area.

cc @vfdev-5

pytorch-bot · 2023-07-11T19:08:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/7735

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 16 New Failures, 14 Unrelated Failures

As of commit 73212ca with merge base f7d9e75 ():

NEW FAILURES - The following jobs have failed:

Build M1 Wheels / pytorch/vision / upload / wheel-py3_8-cpu (gh)
Unable to find any artifacts for the associated workflow
CMake / macos (macos-m1-stable) / macos-job (gh)
The process '/usr/bin/git' failed with exit code 128
CMake / windows (windows.4xlarge, cpu) / windows-job (gh)
The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
CMake / windows (windows.g5.4xlarge.nvidia.gpu, cuda, 11.8) / windows-job (gh)
The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
Docs / build / linux-job (gh)
No files were found with the provided path: /home/ec2-user/actions-runner/_work/_temp/artifacts/. No artifacts will be uploaded.
Tests / unittests-macos (3.10, macos-m1-stable) / macos-job (gh)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-macos (3.11, macos-m1-stable) / macos-job (gh)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-macos (3.12, macos-m1-stable) / macos-job (gh)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-macos (3.8, macos-m1-stable) / macos-job (gh)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-macos (3.9, macos-m1-stable) / macos-job (gh)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-windows (3.10, windows.4xlarge, cpu) / windows-job (gh)
The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
Tests / unittests-windows (3.11, windows.4xlarge, cpu) / windows-job (gh)
The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
Tests / unittests-windows (3.12, windows.4xlarge, cpu) / windows-job (gh)
The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
Tests / unittests-windows (3.8, windows.4xlarge, cpu) / windows-job (gh)
The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
Tests / unittests-windows (3.8, windows.g5.4xlarge.nvidia.gpu, cuda, 11.8) / windows-job (gh)
The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
Tests / unittests-windows (3.9, windows.4xlarge, cpu) / windows-job (gh)
The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

CMake / linux (linux.12xlarge, cpu) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
CMake / linux (linux.g5.4xlarge.nvidia.gpu, cuda, 11.8) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Lint / c-source / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Lint / python-source-and-configs / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Lint / python-types / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Tests / onnx / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-extended / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-linux (3.10, linux.12xlarge, cpu) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-linux (3.11, linux.12xlarge, cpu) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-linux (3.12, linux.12xlarge, cpu) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-linux (3.8, linux.12xlarge, cpu) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-linux (3.8, linux.g5.4xlarge.nvidia.gpu, cuda, 11.8) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128
Tests / unittests-linux (3.9, linux.12xlarge, cpu) / linux-job (gh) (matched linux rule in flaky-rules.json)
The process '/usr/bin/git' failed with exit code 128

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

Build M1 Conda / pytorch/vision / upload / conda-py3_8-cpu (gh) (trunk failure)
Unable to find any artifacts for the associated workflow

This comment was automatically generated by Dr. CI and updates every 15 minutes.

facebook-github-bot · 2023-07-11T19:08:54Z

Hi @antoinebrl!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, 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.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

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

NicolasHug · 2023-07-11T19:12:39Z

Thanks for the PR @antoinebrl , I'm curious about the applications where you found this feature useful?

antoinebrl · 2023-07-11T20:28:31Z

Hi @NicolasHug. Filtering by either size and/or area allows for more control when selecting bounding boxes with different aspect ratios. The smallest edge of a rectangle can be smaller than the sides of a square but the rectangle can occupy more space.

facebook-github-bot · 2023-07-11T20:57:52Z

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

NicolasHug · 2023-07-12T07:16:17Z

Isn't passing min_area=N strictly equivalent to passing min_size=sqrt(N)?

antoinebrl · 2023-07-12T07:39:37Z

If you have square bounding boxes this would be equivalent indeed. However, if there are rectangles with one side ~3-4x longer than the other one, they cover a much larger area than squares with the same small edge. One might want to drop the small squares and keep the rectangles. This effect can be accentuated when resizing images without keeping the aspect ratio.

NicolasHug · 2023-07-12T08:33:53Z

To be clear, min_size will drop a bbox if any of H or W is lower than min_size. So passing min_area=N is equivalent to passing min_size=sqrt(N).

taking N = a * b, the only way for a * b to be lower than N is for either a or b to be less than sqrt(N). I.e. both parametrizations are equivalent. (here, N is min_area and both a and b are min_size)

pmeier · 2023-07-12T09:24:57Z

Here is "proof" of Nicolas statement above

a * b <= N with a, b, N > 0

Assume a = sqrt(N) + eps with eps > 0

(sqrt(N) + eps) * b <= N <=> b <= N / (sqrt(N) + eps) = sqrt(N) * sqrt(N) / (sqrt(N) + eps)

With c = sqrt(N) / (sqrt(N) + eps) and 0 < c < 1

=> b <= sqrt(N) * c < sqrt(N)

antoinebrl · 2023-07-12T21:37:23Z

The min_size only tells one thing about the removed bounding boxes. I made a sketch to illustrate the case I want to cover. In this scenario, how would you drop the square (considered too small) and keep both rectangles (who occupy a much larger region)?

pmeier · 2023-07-13T07:57:28Z

Thanks for the sketch @antoinebrl! Here is a snippet to play with it:

import math

import torch

from torchvision import datapoints
import torchvision.transforms.v2 as transforms
from torchvision.utils import draw_bounding_boxes
from torchvision.ops import box_area
from torchvision.transforms.v2 import functional as F


def draw(bounding_boxes, *, name):
    image = torch.zeros((3, *bounding_boxes.spatial_size), dtype=torch.uint8)
    annotated_image = draw_bounding_boxes(image, bounding_boxes, colors=["red", "green", "blue"])
    F.to_image_pil(annotated_image).save(f"{name}.png")


input = datapoints.BoundingBox(
    [
        [20, 20, 360, 80],  # large rectangle
        [40, 40, 380, 60],  # small rectangle
        [30, 30, 70, 70],  # black square
    ],
    format=datapoints.BoundingBoxFormat.XYXY,
    spatial_size=(100, 400),
)
draw(input, name="input")


areas = box_area(input)
square_area = int(areas[2])

transform = transforms.SanitizeBoundingBox(min_size=math.sqrt(square_area) + 1, labels_getter=None)
draw(transform(input), name="output_min_size")

transform = transforms.SanitizeBoundingBox(min_area=square_area + 1, labels_getter=None)
draw(transform(input), name="output_min_area")

You are right that just min_size there is no way that I see to just drop the square, but not the larger rectangle. What both @NicolasHug and I have neglected above is that we may want to keep bounding boxes that have a shorter side length but ultimately a larger size because of a large aspect ratio.

NicolasHug

I see, thanks for clarifying @antoinebrl . This is indeed not equivalent.

I wonder if we should just accept a filter callable (in addition to min_size which I think we should keep) so that users can pass arbitrary filters. @pmeier ?

Didn't mean to approve just now, sorry (brain fart)

pmeier · 2023-07-13T08:51:55Z

I wonder if we should just accept a filter callable (in addition to min_size which I think we should keep) so that users can pass arbitrary filters.

IMO min_size and min_area are reasonably fundamental to have them as keywords. However, I agree that if anything else comes along we probably should just accept a callable instead of providing more builtin functionality.

NicolasHug · 2023-07-13T08:57:31Z

I'm not convinced TBH. How do we justify min_area over e.g. min_height or min_ratio or min_coverage or min_iou? It seems a bit arbitrary to me for now.

…x-min-area

NicolasHug · 2024-06-04T13:55:43Z

Alright, let's go with min_area. We'll see if users want more filtering strategies. Thanks a lot for the PR @antoinebrl , I took the liberty to resolve some merge conflicts with the latest branch and made some minor updates, but the PR was already in great state.

Summary: Co-authored-by: Nicolas Hug <[email protected]> Reviewed By: vmoens Differential Revision: D58283864 fbshipit-source-id: d598aa25d3d581b1af0f35d082fc050b353488ba

SanitizeBoundingBox based on minimum area

5e5358b

facebook-github-bot added the cla signed label Jul 11, 2023

NicolasHug previously approved these changes Jul 13, 2023

View reviewed changes

Fix tests

f4a2e1e

NicolasHug added 6 commits June 4, 2024 12:34

Merge branch 'main' of github.com:pytorch/vision into ab/sanitize-bbo…

a5a52e3

…x-min-area

Put back tests, also update functional part

5643c40

Lint

e147ca8

Add test, put back shuffle

32842c0

Fix doc

67c9e9e

Merge branch 'main' of github.com:pytorch/vision into ab/sanitize-bbo…

73212ca

…x-min-area

NicolasHug merged commit 1023987 into pytorch:main Jun 4, 2024
9 of 37 checks passed

NicolasHug added enhancement module: transforms labels Jun 4, 2024

This pull request was closed.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

`SanitizeBoundingBox` based on minimum area #7735

`SanitizeBoundingBox` based on minimum area #7735

antoinebrl commented Jul 11, 2023 •

edited by pytorch-bot bot

Loading

pytorch-bot bot commented Jul 11, 2023 •

edited

Loading

facebook-github-bot commented Jul 11, 2023

NicolasHug commented Jul 11, 2023

antoinebrl commented Jul 11, 2023 •

edited

Loading

facebook-github-bot commented Jul 11, 2023

NicolasHug commented Jul 12, 2023

antoinebrl commented Jul 12, 2023

NicolasHug commented Jul 12, 2023 •

edited

Loading

pmeier commented Jul 12, 2023

antoinebrl commented Jul 12, 2023 •

edited

Loading

pmeier commented Jul 13, 2023 •

edited

Loading

NicolasHug left a comment

pmeier commented Jul 13, 2023

NicolasHug commented Jul 13, 2023 •

edited

Loading

NicolasHug commented Jun 4, 2024

SanitizeBoundingBox based on minimum area #7735

SanitizeBoundingBox based on minimum area #7735

Conversation

antoinebrl commented Jul 11, 2023 • edited by pytorch-bot bot Loading

pytorch-bot bot commented Jul 11, 2023 • edited Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/7735

❌ 16 New Failures, 14 Unrelated Failures

facebook-github-bot commented Jul 11, 2023

Action Required

Process

NicolasHug commented Jul 11, 2023

antoinebrl commented Jul 11, 2023 • edited Loading

facebook-github-bot commented Jul 11, 2023

NicolasHug commented Jul 12, 2023

antoinebrl commented Jul 12, 2023

NicolasHug commented Jul 12, 2023 • edited Loading

pmeier commented Jul 12, 2023

antoinebrl commented Jul 12, 2023 • edited Loading

pmeier commented Jul 13, 2023 • edited Loading

NicolasHug left a comment

Choose a reason for hiding this comment

pmeier commented Jul 13, 2023

NicolasHug commented Jul 13, 2023 • edited Loading

NicolasHug commented Jun 4, 2024

`SanitizeBoundingBox` based on minimum area #7735

`SanitizeBoundingBox` based on minimum area #7735

antoinebrl commented Jul 11, 2023 •

edited by pytorch-bot bot

Loading

pytorch-bot bot commented Jul 11, 2023 •

edited

Loading

antoinebrl commented Jul 11, 2023 •

edited

Loading

NicolasHug commented Jul 12, 2023 •

edited

Loading

antoinebrl commented Jul 12, 2023 •

edited

Loading

pmeier commented Jul 13, 2023 •

edited

Loading

NicolasHug commented Jul 13, 2023 •

edited

Loading