TREC Video Retrieval Evaluation

TRECVID 2021


Computer Graphics Computer Vision & Pattern Recognition



TRECVID is the premier annual international workshop event for evaluating content-based retrieval of multimedia and digital video. The workshop organizes a set of different tracks by providing data, task definition, metrics, and evaluation protocols to participants. During the workshop, participants submit notebook papers, discuss results, and benefit from sharing their experiences about which methods work and what does not work and why?!
With its high quality and low cost, it provides an exceptional value for students, academics and industry researchers.
Responses to the call for participation are requested by May 3. All registered teams will be added to a slack workspace by March 1, and all task guidelines will be finalized by April 1.
We look forward to welcome you again as well as new teams and members!
This year TRECVID is running 6 challenge tasks:
1- Ad-hoc Video Search - Given a text query, return the relevant set of videos.
2- Instance Search - Given image examples of a specific person and action, return the person doing the target action.
3- Video to Text - Generate a text caption describing a short (max 10 sec) video. Also, there is a new subtask to fill-in-the-blank space in a sentence that describes a video.
4- Video Summarization - Generate a video summary of major life events for a chosen actor in specific BBC EastEnders set of episodes.
5- Disaster Scene Description and Indexing - Classify scenes after natural disaster events using predefined labels.
6- Activities in Extended Videos - Activity detection from long videos including human and/or object activities from surveillance cameras.
Call for participation is now available:
http://www-nlpir.nist.gov/projects/tv2021/tv21.call.html
Draft version of the TRECVID 2021 tasks guidelines:
http://www-nlpir.nist.gov/projects/tv2021/index.html
Reference: The Scholarly Impact of TRECVid (2003-2009)
http://onlinelibrary.wiley.com/doi/10.1002/asi.21494/full