TRECVID Data Availability

TRECVID 2023

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

• HAVIC dataset for multimedia event detection (through Linguistic Data Consortium (LDC))

The following data is publicly available:

• Internet Archive videos (IACC.3) under Creative Commons licenses (used as development data since 2019). The collection.xml file (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2016-2018. The comment at the top of the collection file tells how to construct the urls for downloading the videos.
• master shot boundary reference for Ad-hoc IACC.3 test data
• ASR output on IACC.3 videos


• The Vimeo Creative Commons (V3C1) dataset under Creative Commons licenses (used as testing data between 2019 - 2021).
  In total, there is 7475 Vimeo videos (1.3 TB, 1000 h) with mean duration of 8 min. All videos have some metadata available
  e.g., title, keywords, and description in json files. The dataset has been segmented into 1,082,657 short video segments
  according to the provided master shot boundary files. In addition, Keyframes and thumbnails per video segment have been extracted and available.
  To download the data as segmented videos please submit a signed data agreement to Angela Ellis. To download the raw whole videos from ITEC University servers please indicate so in your email submitting the form.
  
• The master shot reference for the V3C1 dataset can be downloaded from here with a readme file


• The Vimeo Creative Commons (V3C2) dataset under Creative Commons licenses (used as testing data starting in 2023).
  In total, there is 9760 Vimeo videos (1.6 TB, 1300 h) with mean duration of 8 min. All videos have some metadata available
  e.g., title, keywords, and description in json files. The dataset has been segmented into 1,425,454 short video segments
  according to the provided master shot boundary files. In addition, Keyframes and thumbnails per video segment have been extracted and available.
  To download the data as segmented videos please submit a signed data agreement to Angela Ellis. To download the raw whole videos from ITEC University servers please indicate so in your email submitting the form.
  
• The master shot reference for the V3C2 dataset can be downloaded from here with a readme file


• The Kinolorber movies dataset (MCOLL) is composed of 10 licensed movies used in the Deep Video Understanding (DVU) task. The data is available after submitting a data agreement form found HERE.
• The Deep Video Understanding Movies Training Dataset is a set of 14 Creative Common (CC) movies (total duration of 17.5 hr) previously utilized between 2020 and 2023 ACM Multimedia DVU Grand Challenges and TRECVID DVU including their movie-level and scene-level annotations. The movies have been collected from public websites such as Vimeo and the Internet Archive. In total, the 14 movies consist of 621 scenes, 1572 entities, 650 relationships, and 2491 interactions. The development dataset can be accessed from this URL. Please consult the included documentation folder readme files which explains how the dataset is organized.


• The Video to Text Description (VTT) main task and robustness sub-task testing datasets (V3C3 video shots) is available here
  


• Ground truth data created at/for NIST
• Ad-hoc video search 2023 main task topics (queries)
• Ad-hoc video search truth judgments for 20 main queries
• Ad-hoc video search truth judgments format info


• The Video to Text Description Generation ground truth data


• Deep Video Understanding (DVU) queries and ground truth of the testing dataset

TRECVID 2022

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

• HAVIC dataset for multimedia event detection (through Linguistic Data Consortium (LDC))

The following data is publicly available:

• Internet Archive videos (IACC.3) under Creative Commons licenses (used as development data since 2019). The collection.xml file (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2016-2018. The comment at the top of the collection file tells how to construct the urls for downloading the videos.
• master shot boundary reference for Ad-hoc IACC.3 test data
• ASR output on IACC.3 videos


• The Vimeo Creative Commons (V3C1) dataset under Creative Commons licenses (used as testing data between 2019 - 2021).
  In total, there is 7475 Vimeo videos (1.3 TB, 1000 h) with mean duration of 8 min. All videos have some metadata available
  e.g., title, keywords, and description in json files. The dataset has been segmented into 1,082,657 short video segments
  according to the provided master shot boundary files. In addition, Keyframes and thumbnails per video segment have been extracted and available.
  To download the data as segmented videos please submit a signed data agreement to Angela Ellis. To download the raw whole videos from ITEC University servers please indicate so in your email submitting the form.
  
• The master shot reference for the V3C1 dataset can be downloaded from here with a readme file


• The Vimeo Creative Commons (V3C2) dataset under Creative Commons licenses (used as testing data starting in 2022).
  In total, there is 9760 Vimeo videos (1.6 TB, 1300 h) with mean duration of 8 min. All videos have some metadata available
  e.g., title, keywords, and description in json files. The dataset has been segmented into 1,425,454 short video segments
  according to the provided master shot boundary files. In addition, Keyframes and thumbnails per video segment have been extracted and available.
  To download the data as segmented videos please submit a signed data agreement to Angela Ellis. To download the raw whole videos from ITEC University servers please indicate so in your email submitting the form.
  
• The master shot reference for the V3C2 dataset can be downloaded from here with a readme file


• The Kinolorber movies dataset (MCOLL) is composed of 10 licensed movies and has been used in the Movie Summarization (MSUM) task and the Deep Video Understanding (DVU) task. The data is available after submitting a data agreement form found HERE. The msum testing queries can be downloaded from here
.

• The Deep Video Understanding Movies Training Dataset is a set of 14 Creative Common (CC) movies (total duration of 17.5 hr) previously utilized in 2020 and 2021 ACM Multimedia DVU Grand Challenges including their movie-level and scene-level annotations. The movies have been collected from public websites such as Vimeo and the Internet Archive. In total, the 14 movies consist of 621 scenes, 1572 entities, 650 relationships, and 2491 interactions. The development dataset can be accessed from this URL. Please consult the included documentation folder readme files which explains how the dataset is organized.


• The Video to Text Description (VTT) testing data (V3C1 video shots) is available here
  
• The Disaster Scene Description and Indexing (DSDI) task 6hr testing video shots are available for download. the shots have been collected from various natural disaster events in the USA through FEMA (Federal Emergency Management Agency).


• Ground truth data created at/for NIST
• Ad-hoc video search 2022 main task topics (queries)
• Ad-hoc video search truth judgments for 30 main queries
• Ad-hoc video search truth judgments format info


• The Video to Text Description Generation ground truth data


• Disaster Scene Description and Indexing Task judgments
• Disaster Scene Description and Indexing Task truth judgments format info


• Deep Video Understanding (DVU) queries and ground truth of the testing dataset

TRECVID 2021

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

• HAVIC dataset for multimedia event detection (through Linguistic Data Consortium (LDC))

The following data is publicly available:

• Internet Archive videos (IACC.3) under Creative Commons licenses (used as development data since 2019). The collection.xml file (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2016-2018. The comment at the top of the collection file tells how to construct the urls for downloading the videos.
• master shot boundary reference for Ad-hoc IACC.3 test data
• ASR output on IACC.3 videos


• master shot boundary reference for EastEnders data (Instance Search Task)


• The Vimeo Creative Commons (V3C1) dataset under Creative Commons licenses (used as testing data since 2019).
  In total, there is 7475 Vimeo videos (1.3 TB, 1000 h) with mean duration of 8 min. All videos have some metadata available
  e.g., title, keywords, and description in json files. The dataset has been segmented into 1,082,657 short video segments
  according to the provided master shot boundary files. In addition, Keyframes and thumbnails per video segment have been extracted and available.
  To download the data as segmented videos please submit a signed data agreement to Angela Ellis. To download the raw whole videos from ITEC University servers please indicate so in your email submitting the form.
  
• The master shot reference for the V3C1 dataset can be downloaded from here with a readme file
• The Video Summarization (VSUM) task Queries, Character images, and assessor questions
• The Video to Text Description (VTT) testing data (V3C2 video shots) is available here
• The Disaster Scene Description and Indexing (DSDI) task 6hr original testing videos, segmented shots, corresponding master shot table, and location data are available for download.


• Ground truth data created at/for NIST
• Ad-hoc video search 2021 main task topics (queries)
• Ad-hoc video search truth judgments for 20 main queries
• Ad-hoc video search 2019-2021 progress topics (queries)
• Ad-hoc video search truth judgments for 20 progress queries
• Ad-hoc video search truth judgments format info


• instance search judgments for 20 main task queries
• instance search 2021 main and progress topics (2019-2021)
• Instance search judgments for progress task 2019-2021
• Instance search truth judgments format info


• The Video to Text Description Generation ground truth data
• The fill in the blanks VTT subtask queries and ground truth data


• Disaster Scene Description and Indexing Task judgments
• Disaster Scene Description and Indexing Task truth judgments format info

TRECVID 2020

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

• HAVIC dataset for multimedia event detection (through Linguistic Data Consortium (LDC))

The following data is publicly available:

• Internet Archive videos (IACC.3) under Creative Commons licenses (used as development data in 2019). The collection.xml file (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2016-2018. The comment at the top of the collection file tells how to construct the urls for downloading the videos.
• master shot boundary reference for Ad-hoc IACC.3 test data
• ASR output on IACC.3 videos


• master shot boundary reference for EastEnders data (Instance Search Task)


• The Vimeo Creative Commons (V3C1) dataset under Creative Commons licenses (used as testing data in 2019).
  In total, there is 7475 Vimeo videos (1.3 TB, 1000 h) with mean duration of 8 min. All videos have some metadata available
  e.g., title, keywords, and description in json files. The dataset has been segmented into 1,082,657 short video segments
  according to the provided master shot boundary files. In addition, Keyframes and thumbnails per video segment have been extracted and available.
  To download the data as segmented videos please submit a signed data agreement to Angela Ellis. To download the raw whole videos from ITEC University servers please indicate so in your email submitting the form.
  
• The master shot reference for the V3C1 dataset can be downloaded from here with a readme file
• The Video to Text Description (VTT) testing data (V3C2 video shots) and associated text description subsets are available here
• The Disaster Scene Description and Indexing (DSDI) task 5hr original testing videos, segmented shots, and corresponding master shot table are available for download.
• The Video Summarization (VSUM) task Queries, Character images, assessor questions, Results, and submission XMLS are available for download.


• Truth data created at/for NIST
• Ad-hoc video search topics (queries)
• Ad-hoc video search truth judgments for 20 queries used for the official workshop scoring
• Ad-hoc video search truth judgments format info


• instance search judgments
• Instance search truth judgments format info
• instance search topics and included images


• The Video to Text Description Generation ground truth data
• The Video to Text Matching & Ranking ground truth data


• Disaster Scene Description and Indexing Task judgments
• Disaster Scene Description and Indexing Task truth judgments format info

TRECVID 2019

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection (through Linguistic Data Consortium (LDC))

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.3) under Creative Commons licenses (used as development data in 2019). The collection.xml file (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2016-2018. The comment at the top of the collection file tells how to construct the urls for downloading the videos.
• master shot boundary reference for Ad-hoc test data
• ASR output on IACC.3 videos


• master shot boundary reference for EastEnders data (Instance Search Task)


• The Vimeo Creative Commons (V3C1) dataset under Creative Commons licenses (used as testing data in 2019).
  In total, there is 7475 Vimeo videos (1.3 TB, 1000 h) with mean duration of 8 min. All videos have some metadata available
  e.g., title, keywords, and description in json files. The dataset has been segmented into 1,082,657 short video segments
  according to the provided master shot boundary files. In addition, Keyframes and thumbnails per video segment have been extracted and available.
  To download the data as segemnted videos please submit a signed data agreement to Angela Ellis. To downlaod the raw whole videos from ITEC University servers please indicate so in your email submitting the form.
  
• The master shot reference for the V3C1 dataset can be downloaded from here with a readme file
• The Video to Text Description testing data (Vine URLs) and associated text description subsets are available here


• Truth data created at/for NIST
• Ad-hoc video search topics (queries)
• Ad-hoc video search truth judgments for 30 queries used for the official workshop scoring
• Ad-hoc video search truth judgments format info


• instance search judgments
• Instance search truth judgments format info
• instance search topics and included images


• The Video to Text Description Generation ground truth data


• The Video to Text Matching & Ranking ground truth data

TRECVID 2018

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection (through Linguistic Data Consortium (LDC))

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.3) under Creative Commons licenses. The collection.xml file (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2016-2018. The comment at the top of the collection file tells how to construct the urls for downloading the videos.
• master shot boundary reference for Ad-hoc test data
• ASR output on IACC.3 videos


• master shot boundary reference for EastEnders data (Instance Search Task)


• The Video to Text Description testing data (Vine URLs) and associated text description subsets are available here


• Truth data created at/for NIST
• Ad-hoc video search topics (queries)
• Ad-hoc video search truth judgments for 30 queries used for the official workshop scoring
• Ad-hoc video search truth judgments format info


• instance search judgments
• instance search topics and included images


• The Video to Text Description Generation ground truth data


• The Video to Text Matching & Ranking ground truth data

TRECVID 2017

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.3) under Creative Commons licenses. The collection.xml file (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2016-2017. The comment at the top of the collection file tells how to construct the urls for downloading the videos.
• master shot boundary reference for Ad-hoc test data
• ASR output on IACC.3 videos


• master shot boundary reference for EastEnders data (Instance Search Task)


• The Video to Text Description testing data (Vine URLs) and associated text description subsets are available here


• Truth data created at/for NIST
• Ad-hoc video search topics (queries)
• Ad-hoc video search truth judgments for 30 queries used for the official workshop scoring
• Ad-hoc video search truth judgments format info


• instance search judgments
• instance search topics and included images


• The Video to Text Description Generation ground truth data


• The Video to Text Matching & Ranking ground truth data

TRECVID 2016

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.3) under Creative Commons licenses. The collection.xml file (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2016. The comment at the top of the collection file tells how to construct the urls for downloading the videos.
• master shot boundary reference for Ad-hoc test data
• ASR output on IACC.3 videos
• Ad-hoc video search topics (queries)


• The localization I-frames used in 2016 are composed of the I-frames of IACC.2.A-C:
  The localization I-frames for the IACC.2.A dataset
  The localization I-frames for the IACC.2.B dataset
  The localization I-frames for the IACC.2.C dataset


• The Video to Text Description testing data (Vine URLs) and associated two text description sets are available here


• Truth data created at/for NIST
• Ad-hoc video search truth judgments for 30 queries used for the official workshop scoring
• Ad-hoc video search truth judgments format info
• Extra Ad-hoc video search judgments made by VBS and Klagenfurt University


• localization testing data
• localization truth data header info
• localization truth data


• instance search judgments
• master shot boundary reference for EastEnders data
• instance search topics and included images


• Video to Text ground truth header info
• The Video to Text Description ground truth data

TRECVID 2015

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.2) under Creative Commons licenses

  The collection.xml files (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2013, 14, and 15. The comment at the top of the collection file tells how to construct the urls for checking the licensing and downloading the video. Note that some files are included in collection.xml but have their USE marked as dropped due to a change in their status at the Internet Archive.

• The localization I-frames for the IACC.2.C dataset


• Truth data created at/for NIST
• semantic indexing truth judgments for 30 single concepts used for the official workshop scoring
• semantic indexing truth judgments format info
• localization truth data header info
• localization truth data
• instance search judgments
• master shot boundary reference for semantic indexing test data
• master shot boundary reference for EastEnders data
• instance search topics and included images
• ASR output on IACC.2 videos

TRECVID 2014

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.2) under Creative Commons licenses

  The collection.xml files (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2013, 14, and 15. The comment at the top of the collection file tells how to construct the urls for checking the licensing and downloading the video. Note that some files are included in collection.xml but have their USE marked as dropped due to a change in their status at the Internet Archive.

• The localization I-frames for the IACC.2.B dataset


• Truth data created at/for NIST
• semantic indexing truth judgments for 30 single concepts used for the official workshop scoring
• semantic indexing truth judgments format info
• localization truth data header info
• localization truth data
• instance search judgments
• master shot boundary reference for semantic indexing test data
• master shot boundary reference for EastEnders data
• instance search topics and included images
• ASR output on IACC.2 videos

TRECVID 2013

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007-9 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.2) under Creative Commons licenses

  The collection.xml files (load slowly) contains urls for the Internet Archive videos for use in TRECVID 2013, 14, and 15. The comment at the top of the collection file tells how to construct the urls for checking the licensing and downloading the video. Note that some files are included in collection.xml but have their USE marked as dropped due to a change in their status at the Internet Archive.

• The localization I-frames for the IACC.2.A dataset


• Truth data created at/for NIST
• semantic indexing truth judgments for 38 single concepts used for the official workshop scoring
• semantic indexing truth judgments for 60 single concepts (provided post-workshop using additional judgments by QUAERO)
• semantic indexing truth judgments for 10 concept pairs (basis for workshop scoring, by NIST and QUAERO)
• localization truth data header info
• localization truth data
• instance search judgments
• master shot boundary reference for semantic indexing test data
• master shot boundary reference for EastEnders data
• instance search topics and included images
• ASR output on IACC.2 videos

TRECVID 2012

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2009 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.1) under Creative Commons licenses

  The iacc.1.C.collection.xml file (loads slowly) contains urls for the Internet Archive videos used in TRECVID 2012. The comment at the top of the collection file tells how to construct the urls for checking the licensing and downloading the video. Note that some files are included in collection.xml but marked as dropped due to a change in their status at the Internet Archive.

• Truth data created at/for NIST
• semantic indexing truth judgments for 46 single concepts (basis for workshop scoring, by NIST and QUAERO)
• semantic indexing truth judgments for 10 concept pairs (basis for workshop scoring, by NIST and QUAERO)
• known-item search groundtruth
• instance search judgments
• Truth data and baselines created by participants
• Common annotations for the semantic indexing data
• master shot boundary reference for semantic indexing training and test data
• known-item test topics (automatic)
• known-item test topics (interactive)
• instance search topics and included images
• ASR output on IACC.1.C videos

TRECVID 2011

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2009 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.1) under Creative Commons licenses

  The iacc.1.B.collection.xml file (loads slowly) contains urls for the Internet Archive videos used in TRECVID 2011. The comment at the top of the collection file tells how to construct the urls for checking the licensing and downloading the video. Note that some files are included in collection.xml but marked as dropped due to a change in their status at the Internet Archive.

• Truth data created at/for NIST
• semantic indexing truth judgments for 50 concepts (basis for workshop scoring, by NIST and QUAERO)
• known-item search groundtruth
• instance search judgments
• copy detection groundtruth and a+v query script
• Truth data and baselines created by participants
• Common annotations, models for semantic indexing data
• master shot boundary reference for semantic indexing training and test data
• known-item test topics (automatic)
• known-item test topics (interactive)
• instance search topics and included images
• ASR output on IACC.1.B videos

TRECVID 2010

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2009 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection
• HAVIC dataset for multimedia event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Internet Archive videos (IACC.1) under Creative Commons licenses

  The iacc.1.A.collection.xml file (loads slowly) contains urls for the Internet Archive videos used in TRECVID 2010. The comment at the top of the collection file tells how to construct the urls for checking the licensing and downloading the video. Note that some files are included in A.collection.xml but marked as dropped due to a change in their status at the Internet Archive.

• Truth data created at/for NIST
• semantic indexing truth judgments for 30 concepts (basis for workshop scoring, by NIST and QUAERO)
• semantic indexing truth judgments for 20 additional concepts (by QUAERO)
• known-item search groundtruth
• instance search judgments
• copy detection groundtruth and a+v query script
• Truth data and baselines created by participants
• Common annotations, models for semantic indexing data
• master shot boundary reference for semantic indexing training and test data
• master shot boundary reference for instance search test data (tv9.sv.test)
• known-item sample topics with ground truth
• known-item test topics
• instance search topics and included images
• ASR output on IACC.1.A and training videos

TRECVID 2009

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2009 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2009

The following data is only available from NIST for current TRECVID participants:

• BBC rushes video files (MPEG-1)
• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Truth data created at/for NIST
• high-level feature truth judgments
• search relevance judgments
• copy detection groundtruth and a+v query script
• Truth data and baselines created by participants
• Common annotations, models for feature/search data
• master shot boundary reference for feature/search data
• search topics and included images
• ASR output on feature/search data
• MT output on feature/search data

TRECVID 2008

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2008 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2008

The following data is available from NIST for current TRECVID participants:

• BBC rushes video files (MPEG-1)
• Development dataset annotations for surveillance event detection
• Test dataset annotations for surveillance event detection

• London Gatwick surveillance video files

The i-LIDS Multiple Camera Tracking Scenario Training data set is the original source of the evaluation set for TRECVID SED 2009, 2010, and 2011. It is available for non-TRECVID participants through the standard i-LIDS licensing process

The following data is publicly available:

• Truth data created at/for NIST
• high-level feature truth judgments
• search relevance judgments
• video summary test inclusion groundtruth
• video summary full inclusion groundtruth
• copy detection groundtruth and a+v query script
• Truth data and baselines created by participants
• Common annotations for feature/search data

The annotations for the event detection test data are being withheld pending decisions about plans for TRECVID 2009.

• master shot boundary reference for feature/search data
• search topics and included images
• ASR output on feature/search data
• MT output on feature/search data

TRECVID 2007

The following data is available for multimedia research only, under restrictions agreed to with the Netherlands Institute for Sound and Vision. The data can be requested by filling out and returning the forms available here.

• TRECVID 2007 Sound and Vision video files (MPEG-1)
• LIG keyframes distributed in 2007

The following data is only available from NIST for current TRECVID participants:

• BBC rushes video files (MPEG-1)

The following data is publicly available:

• Truth data created at/for NIST
• shot boundary annotation
• high-level feature truth judgments
• search relevance judgments
• video summary test inclusion groundtruth (from DCU)
• video summary full inclusion groundtruth (DCU)
• Truth data and baselines created by participants
• Common annotations
• master shot boundary reference
• search topics and included images
• ASR output
• MT output

TRECVID 2006

All or part of the following data may be available for purchace from the Linguistic Data Consortium. Check the LDC catalog. NIST is not able to make the full video available.

• broadcast news video files (MPEG-1)
• master keyframes

The following data is publicly available:

• Truth data created at NIST
• shot boundary annotation
• high-level feature truth judgments
• search relevance judgments
• Truth data and baselines created by participants
• MediaMill Baseline
• Columbia University's Baseline Detectors for 374 LSCOM Semantic Visual Concepts
• LSCOM annotations
• master shot boundary reference
• search topics and included images
• ASR/MT output

TRECVID 2005

All or part of the following data may be available for purchace from the Linguistic Data Consortium. Check the LDC catalog. NIST is not able to make the full video available.

• broadcast news video files (MPEG-1)
• master keyframes selected by Dublin City University

The following data is publicly available:

• Truth data created at NIST
• shot boundary annotation
• low-level feature truth judgments
• high-level feature truth judgments
• search relevance judgments
• Truth data created by participants
• camera motion annotation donated by Joanneum and KDDI
• common development feature annotation using the CMU and IBM tools
• low-level development features donated by CMU
• Tsinghua - Intel object annotations
• master shot boundary reference
• search topics and included images
• ASR/MT output

TRECVID 2004

All or part of the following data may be available for purchace from the Linguistic Data Consortium. Check the LDC catalog. NIST is not able to make the full video available.

• video files (MPEG-1)
• master keyframes selected by CLIPS-IMAG
• ASR output created by LIMSI
• truth data created at LDC
• story boundary annotation

The following data is publicly available:

• Truth data created at NIST
• feature truth judgments
• search relevance judgments
• shot boundary annotation
• Truth data created by participants
• master shot boundary reference
• search topics and included images

TRECVID 2003

All or part of the following data may be available for purchace from the Linguistic Data Consortium. Check the LDC catalog. NIST is not able to make the full video available.

• video files (MPEG-1)
• master keyframes selected by CLIPS-IMAG
• ASR output created by LIMSI
• truth data created at LDC
• story boundary annotation

The following data is publicly available:

• Truth data created at NIST
• feature truth judgments
• search relevance judgments
• shot boundary annotation
• Truth data created by participants
• common feature annotation
• shared features extracted automatically
• master shot boundary reference
• search topics and included images

TREC 2002 Video Track

• video files (MPEG-1)

Download the OpenVideo files from the Open Video Project using the list of video names in the guidelines. Download the Internet Archive files from the Movie Archive using the list of video names in the guidelines. There maybe some problems in downloading because files have been moved or renamed.

• keyframes selected by Dublin City University
• ASR output created by LIMSI
• Truth data created at NIST
• feature truth judgments
• search relevance judgments
• shot boundary annotation
• shared features extracted automatically
• master shot boundary reference
• search topics

TREC 2001 Video Track

• video files (MPEG-1)

The BBC data is no longer available. Download the OpenVideo files from the Open Video Project using list of video names in the guidelines. There maybe some problems in downloading because files have been moved or renamed. Buy NIST videos from the NIST Standard Reference Data site.

• search topics


• Truth data created at NIST
• shot boundary annotation