Skip to main content

Ego4D Challenge 2022

VQ Test Annotations Released

The VQ test annotations have been release for challenge submissions. Please update your local annotations to obtain them via the CLI, e.g. python -m ego4d.cli.cli --output_directory="~/ego4d_data" --datasets annotations

Overview

In 2022, we will host 16 challenges, representing each of Ego4D’s five benchmarks. These are:

Episodic memory:

  • Visual queries with 2D localization and VQ 3D localization: Given an egocentric video clip and an image crop depicting the query object, return the last time the object was seen in the input video, in terms of the tracked bounding box (2D + temporal localization) or the 3D displacement vector from the camera to the object in the environment.
  • Natural language queries: Given a video clip and a query expressed in natural language, localize the temporal window within all the video history where the answer to the question is evident.
  • Moments queries: Given an egocentric video and an activity name (e.g., a “moment”), localize all instances of that activity in the past video

Hands and Objects:

Audio-Visual Diarization & Social:

  • Audio-visual localization: Given an egocentric video clip, localize the speakers in the visual field of view.
  • Audio-visual speaker diarization: Given an egocentric video clip, identify which person spoke and when they spoke.
  • Audio-only Diarization Challenge: Given an egocentric video clip, identify which person spoke and when they spoke based on audio alone.
  • Speech transcription: Given an egocentric video clip, transcribe the speech of each person.
  • Talking to me: Given an egocentric video clip, identify whether someone in the scene is talking to the camera wearer.
  • Looking at me: Given an egocentric video clip, identify whether someone in the scene is looking at the camera wearer.

Forecasting:

  • Locomotion forecasting: Given a video frame and the past trajectory, predict the future ego positions of the camera wearer (in the form of a 3D trajectory).
  • Hand forecasting: Given a short preceding video clip, predict where the hand will be visible in the future, in terms of a bounding box center in keyframes.
  • Short-term hand object prediction: Given a video clip, predict the next active objects, the next action, and the time to contact.
  • Long-term activity prediction: Given a video clip, the goal is to predict what sequence of activities will happen in the future? For example, after kneading dough, what will the baker do next?

Dataset

Ego4D challenge participants will use Ego4D’s annotated data set of more than 3,670 hours of video data, capturing the daily-life scenarios of more than 900 unique individuals from nine different countries around the world. Unique train, validation and unannotated test sets are available to download per challenge at https://ego4d-data.org/docs/.

Participation Guidelines

Participate in the contest by registering on the EvalAI challenge page and create a team. All participants must register as a part of a “participating team” on EvalAI to ensure the submission limits are honored. Participants will upload their predictions in the format specified for the specific challenge, and will be evaluated on AWS instance by comparing to ground truth predictions. Instructions for training, local evaluation, and online submission are provided at EvalAI. Please refer to the individual EvalAI pages for each challenge for submission guidelines, task specifications, and evaluation criteria.

Dates

Top performing teams will be invited to speak at our accepted CVPR and ECCV workshops in June and October, with the specific competition tracks below concluding respectively at these times.

June 1, 2022 (CVPR Workshop, phase 1)

  • Visual Queries
  • Natural Language Queries
  • PNR Temporal Localization
  • Object State Change Classification
  • Short-term Object Interaction Anticipation
  • Long-term Action Anticipation

October 1, 2022 (ECCV Workshop, phase 2)

  • Moments Queries
  • 3D Localization
  • State Change Object Detection
  • Localization and Tracking
  • Diarization (Audio)
  • Diarization (Audio+Video)
  • Transcription
  • Looking at Me
  • Talking to Me
  • Future Hand Prediction

Competition Rules and Prize Information

Competition rules can be found here. Additionally, we are thrilled that FAIR is able to offer the following prize thresholds per challenges:

  • First place: $3,000
  • Second place: $2,000
  • Third place: $1,000

Challenge Reports

In addition to the submission on EvalAI, participants must submit a report describing their method to the workshop CMT (link TBD). In addition to your method and results, please remember to include examples of positive and negative results (limitations) of your model.

For CVPR'22 challenges due June 1, the deadline for report submission will be June 6.

Acknowledgements

The Ego4D challenge would not have been possible without the infrastructure and support of the EvalAI team. Thank you!

Organizers

Rohit Girdhar

Santhosh Kumar Ramakrishnan

Chen Zhao

Merey Ramazanova

Satwik Kottur

Mengmeng Xu

Vincent Cartillier

Yifei Huang

Qichen Fu

Siddhant Bansal

Hao Jiang

Vamsi Ithapu

Jachym Kolar

Christian Fuegen,

Leda Sari

Eric Zhongcong Xu

Yunyi Zhu

Murong Ma

Zachary Chavis

Wenqi Jia

Miao Liu

Antonino Furnari

Tushar Nagarajan

Karttikeya Mangalam

Dima Damen

Giovanni Maria Farinella

Michael Wray

Gene Byrne

Andrew Westbury