From roddy.collins at kitware.com  Mon Jul 24 13:39:19 2023
From: roddy.collins at kitware.com (Roddy Collins)
Date: Mon, 24 Jul 2023 09:39:19 -0400
Subject: [Viratdata] Inquiry About VIRAT Dataset Annotations for Video
 Object Detection
In-Reply-To: <CAMzEAatfvjPyrS+f-J=vWfJSuVmdM3ApQUBVWNmJ2Tdnp61p=Q@mail.gmail.com>
References: <CAMzEAatfvjPyrS+f-J=vWfJSuVmdM3ApQUBVWNmJ2Tdnp61p=Q@mail.gmail.com>
Message-ID: <CAO=zZiOr52GCGTa-wfME+BKqCFeh6y0856JRdiyj1J-EScq_iw@mail.gmail.com>

Hello,

I believe you are referring to the VIRAT Release 2.0 annotations. The VIRAT
annotations were refreshed for the IARPA DIVA program, as described at
viratdata.org; although DIVA focused on activity recognition rather than
object recognition, I believe the object counts are higher.  We currently
have no projects supporting extensive review of the annotations, although
we welcome submissions of corrected annotations via the git repository.

We also have an additional dataset at mevadata.org; although this is also
focused on activity recognition, objects and props participating in
activities are annotated and may be useful for training.

I hope this helps; let us know if you have any other questions.

Thanks
-Roddy




On Sat, Jul 22, 2023 at 7:35?AM Parisa Ab Savaran <parisaabsavaran at gmail.com>
wrote:

> Dear VIRAT Dataset Team,
>
> We hope this email finds you well. We are currently working on video
> object detection using the VIRAT dataset and have noticed some potential
> issues with the bounding box annotations in certain videos. We would like
> to bring specific instances to your attention:
>
> In the video VIRAT_S_000006, the bounding boxes for humans appear to be
> missing from the annotations.
>
> In videos VIRAT_S_000203_04_000903_001086 and
> VIRAT_S_000207_04_000902_000934, not all cars have been annotated with
> bounding boxes. For instance, "VIRAT_S_000207_04_000902_000934" seems to
> have limited bounding box annotations for objects and people.
>
> In the video VIRAT_S_010202_01_000055_000147, all individuals except the
> person with the bicycle have been included in the annotations.
>
> As video object detection is a critical aspect of our research, having
> access to accurate and comprehensive annotations is crucial for achieving
> meaningful results. We are particularly interested in understanding if the
> VIRAT dataset is fully annotated for video object detection tasks or if
> there are any areas that require improvement.
>
> Thank you for your attention to this matter. I look forward to your
> response and any assistance you can provide.
>
> Best regards,
>
> Vision team of Amirkabir university
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/viratdata/attachments/20230724/58fcc543/attachment.htm>