divided by 0 error

[AlexWang1900]

Before submitting a bug report, please be aware that your issue must be reproducible with all of the following, otherwise it is non-actionable, and we can not help you:

• Current repo: run git fetch && git status -uno to check and git pull to update repo
• Common dataset: coco.yaml or coco128.yaml
• Common environment: Colab, Google Cloud, or Docker image. See https://github.com/ultralytics/yolov5#reproduce-our-environment

If this is a custom dataset/training question you must include your train*.jpg, test*.jpg and results.png figures, or we can not help you. You can generate these with utils.plot_results().

🐛 Bug

in utils.py line 345:

elif CIoU:  # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
                v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / h2) - torch.atan(w1 / h1), 2)
                with torch.no_grad():
                    alpha = v / (1 - iou + v)   ### should add + 1e-16 to avoid alpha = NAN.
                return iou - (rho2 / c2 + v * alpha)  # CIoU

To Reproduce (REQUIRED)

it's hard to reproduce, I encounter it when I was trying some new target assigning methods..

Output:
alpha = Nan.

Expected behavior

alpha be a normal value between 0 - 1

Environment

If applicable, add screenshots to help explain your problem.

• OS: [e.g. Ubuntu]
• GPU [e.g. 2080 Ti]

Additional context

Add any other context about the problem here.
it's hard to reproduce, I encounter it when I was trying some new target assigning methods..

[glenn-jocher]

@AlexWang1900 ah, thanks for the bug report! It looks like c2 is already protected, so then alpha may be the only unprotected divide as you say. Yes we should add 1e-16 here. Would you like to submit a PR for this or I can include it in the next commit?

[AlexWang1900]

I will summit a PR thanks!😀

…

---Original--- From: "Glenn Jocher"<notifications@github.com&gt; Date: Wed, Jul 29, 2020 02:26 AM To: "ultralytics/yolov5"<yolov5@noreply.github.com&gt;; Cc: "Mention"<mention@noreply.github.com&gt;;"AlexWang1900"<41888506@qq.com&gt;; Subject: Re: [ultralytics/yolov5] divided by 0 error (#541) @AlexWang1900 ah, thanks for the bug report! It looks like c2 is already protected, so then alpha may be the only unprotected divide as you say. Yes we should add 1e-16 here. Would you like to submit a PR for this or I can include it in the next commit? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[glenn-jocher]

Issue resolved, removing TODO.