From jim.keeting at gmail.com Wed Sep 15 12:11:40 2021 From: jim.keeting at gmail.com (Jim Keeting) Date: Wed, 15 Sep 2021 13:11:40 +0100 Subject: [Viratdata] VIRAT Dataset clarifications Message-ID: Just a few questions I'd be grateful for any support with. 1. I saw VIRAT Video Dataset Release 2.0/VIRAT Ground Dataset/annotations/VIRAT_S_000003.viratdata.objects.txt has objects of type "0" what does this refer to, as the documentation suggests the objects types are 1-7? 2. For a given event or frame, are all people always labelled in the scene, or just a subselection relevant to the event? Was wondering if there is any way to select frames where all people have been annotated. 3. To download the dataset I tried the MIDAS software and server with: http://midas.kitware.com/api/rest URL but it rejected authentication despite having an account. I can download via the website at https://data.kitware.com/ but cannot find an easy way to download the whole dataset as if one downloads the whole connection it results in the connection terminating before the download is completed or a corrupted .7z archive, whereas I can manually download them one-by-one but there are too many items to download one by one. Thank you, Jim -------------- next part -------------- An HTML attachment was scrubbed... URL: From roddy.collins at kitware.com Tue Sep 21 20:32:30 2021 From: roddy.collins at kitware.com (Roddy Collins) Date: Tue, 21 Sep 2021 16:32:30 -0400 Subject: [Viratdata] VIRAT Dataset clarifications In-Reply-To: References: Message-ID: Hi Jim-- Responses inline. On Wed, Sep 15, 2021 at 8:11 AM Jim Keeting wrote: > 1. I saw VIRAT Video Dataset Release 2.0/VIRAT Ground Dataset/annotations/VIRAT_S_000003.viratdata.objects.txt > has objects of type "0" what does this refer to, as the documentation > suggests the objects types are 1-7? > > Having looked at the objects in question, they seem to be static boxes that attempt to mark out the background -- they roughly mark out the grass, asphalt, sidewalk, and building. My guess is that they were a preliminary attempt at representing background classes that never came to fruition and were not removed from the dataset before release. > 2. For a given event or frame, are all people always labelled in the > scene, or just a subselection relevant to the event? Was wondering if there > is any way to select frames where all people have been annotated. > The VIRAT ground camera data has been annotated twice: once in 2010 as part of the original CVPR paper, and again in 2017 for the IARPA DIVA program. In the original CVPR paper, moving objects were tracked, and associated with any of the VIRAT activities that occurred during motion. (In other words, they might be annotated as movers without being associated with other activities.) If they were stationary, they were not annotated. These are the annotations in the Release 2.0 directory. For the DIVA effort (on VIRAT data), persons and vehicles were tracked all the time, whether or not they were moving or stationary. These annotations are available at https://gitlab.kitware.com/viratdata/viratannotations . (Note that this annotation approach differs from the DIVA annotations on the MEVA data at mevadata.org, where we only annotated actors who were participating in defined DIVA events.) > > 3. To download the dataset I tried the MIDAS software and server with: > http://midas.kitware.com/api/rest URL but it rejected authentication > despite having an account. I can download via the website at > https://data.kitware.com/ but cannot find an easy way to download the > whole dataset as if one downloads the whole connection it results in the > connection terminating before the download is completed or a corrupted .7z > archive, whereas I can manually download them one-by-one but there are too > many items to download one by one. > > I believe this is due to the mechanisms ensuring downloaders have agreed to the ground camera data usage agreement. One potential workaround is to create and upload fewer-but-larger tarballs of the videos; I'll look into this and update when I have more information. I hope this helps; let us know if you have any other questions. Thanks -Roddy -- Roddy Collins, PhD Principal Engineer, Kitware Inc. 1712 Route 9, Suite 300, Clifton Park, NY 12065 (518) 881-4918 || roddy.collins at kitware.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From jim.keeting at gmail.com Mon Sep 27 19:13:03 2021 From: jim.keeting at gmail.com (Jim Keeting) Date: Mon, 27 Sep 2021 20:13:03 +0100 Subject: [Viratdata] VIRAT Dataset clarifications In-Reply-To: References: Message-ID: Thank you that is very helpful. Please keep me informed if you find a way to download the dataset easily these days. Kind regards, Jim On Tue, 21 Sept 2021 at 21:32, Roddy Collins wrote: > Hi Jim-- > > Responses inline. > > On Wed, Sep 15, 2021 at 8:11 AM Jim Keeting wrote: > >> 1. I saw VIRAT Video Dataset Release 2.0/VIRAT Ground Dataset/annotations/VIRAT_S_000003.viratdata.objects.txt >> has objects of type "0" what does this refer to, as the documentation >> suggests the objects types are 1-7? >> >> > Having looked at the objects in question, they seem to be static boxes > that attempt to mark out the background -- they roughly mark out the grass, > asphalt, sidewalk, and building. My guess is that they were a preliminary > attempt at representing background classes that never came to fruition and > were not removed from the dataset before release. > > >> 2. For a given event or frame, are all people always labelled in the >> scene, or just a subselection relevant to the event? Was wondering if there >> is any way to select frames where all people have been annotated. >> > > The VIRAT ground camera data has been annotated twice: once in 2010 as > part of the original CVPR paper, and again in 2017 for the IARPA DIVA > program. In the original CVPR paper, moving objects were tracked, and > associated with any of the VIRAT activities that occurred during motion. > (In other words, they might be annotated as movers without being associated > with other activities.) If they were stationary, they were not annotated. > These are the annotations in the Release 2.0 directory. > > For the DIVA effort (on VIRAT data), persons and vehicles were tracked all > the time, whether or not they were moving or stationary. These annotations > are available at https://gitlab.kitware.com/viratdata/viratannotations . > (Note that this annotation approach differs from the DIVA annotations on > the MEVA data at mevadata.org, where we only annotated actors who were > participating in defined DIVA events.) > > >> >> 3. To download the dataset I tried the MIDAS software and server with: >> http://midas.kitware.com/api/rest URL but it rejected authentication >> despite having an account. I can download via the website at >> https://data.kitware.com/ but cannot find an easy way to download the >> whole dataset as if one downloads the whole connection it results in the >> connection terminating before the download is completed or a corrupted .7z >> archive, whereas I can manually download them one-by-one but there are too >> many items to download one by one. >> >> > I believe this is due to the mechanisms ensuring downloaders have agreed > to the ground camera data usage agreement. One potential workaround is to > create and upload fewer-but-larger tarballs of the videos; I'll look into > this and update when I have more information. > > I hope this helps; let us know if you have any other questions. > > Thanks > -Roddy > > -- > Roddy Collins, PhD > Principal Engineer, Kitware Inc. > 1712 Route 9, Suite 300, Clifton Park, NY 12065 > (518) 881-4918 || roddy.collins at kitware.com > > -------------- next part -------------- An HTML attachment was scrubbed... URL: