Qualcomm Exercise Video Dataset (QEVD)
Qualcomm Exercise Video Dataset (QEVD)
chip image

Qualcomm AI Research presents the Qualcomm Exercise Video Dataset (QEVD) which explores human-AI interaction in the challenging real-world domain of fitness coaching – a task which intrinsically requires monitoring live user activity and providing timely feedback. Our dataset includes corrective feedbacks to address potential user mistakes and steer them towards successful workout completion.

Our dataset contains 474+ hours of videos and includes the following:

  • short-clip videos (∼5 seconds in length) annotated with 1M+ question-answer pairs
  • short-clip videos (∼5 seconds in length) annotated with 650k+ live feedbacks including corrective feedbacks
  • long-range videos (>3 minutes in length) annotated with 7.5k+ live feedbacks including corrective feedbacks

Our Qualcomm Exercise Video Dataset contains three subsets:

  • QEVD-Fit-300k (fine-grained short videos)
    • The QEVD FIT 300K dataset includes 289k video clips between 2 and 10 seconds long of humans performing pre-defined exercises including specific variations of each. The dataset also includes 14k video clips of everyday human activities relevant in a fitness context. The dataset is designed for training machine learning models for fine-grained human motion understanding in the fitness domain.
    • They cover 148 different exercises and their variations including: varied pacing, performing common mistakes, and modified form.
    • Approx. 10 variations were collected per exercise.
  • QEVD-Fit-Coach
    • The QEVD FIT-COACH Dataset includes 149 workout videos of humans performing a structured workout consisting of 4-6 exercises.
    • Each video is ~3.5 minutes long and is annotated with feedbacks from the perspective of a fitness coach.
    • This dataset also includes additional annotations for the QEVD-FIT-300K dataset including:
      • feedbacks from a fitness coach perspective
      • high level questions (e.g., What exercise is the user doing?)
      • fine-grained questions (e.g., Is the user moving their arms?). The dataset is designed for training machine learning models for interactive fitness coaching.
  • QEVD-Fit-Coach-Benchmark
    • The QEVD FIT-COACH Benchmark includes 74 workout videos of humans performing a structured workout consisting of 4-6 exercises.
    • Note: These videos contain distinct participants from the QEVD-FIT-Coach Dataset.
    • Each video is ~3.5 minutes long and is annotated with feedbacks from the perspective of a fitness coach.
    • The benchmark is designed for testing machine learning models for interactive fitness coaching.

Classes

Full list of exercises and fine-grained attributes for the QEVD-FIT-300K video collection. Each video is labeled with one or more label in the format: "<exercise> - <fine-grained attribute>".
 
Please download this file:
  • QEVD-FIT-300K Classes

Example of exercise class and fine-grained attributes

 

Exercise Fine-grained Attributes
Arms moving too much arms flapping too much
flapping
heels only
height=1
height=2
height=3
height=4
height=5
no jump
no obvious issue
no wrist circles
not moving
side to side
speed=0.75 rps
speed=1.00 rps
speed=1.25 rps
speed=1.50 rps
speed=1.75 rps
speed=2.0 rps

 

Example of General actions

General Actions
bobbing head (imagine there is music) neck-warm-up (without hands)
boxing bounce-steps nodding head to say yes (long)
catching your breach (crouching) nodding head to say yes (short)
catching your breath (hand on knees)

open and drink from a
catching your breath (hands behind head) picking up the camera
catching your breath (leaning on something) plank preparation
catching your breath (walking around) pretending to towel off sweat (without using a towel)
changing the webcam view while lying down scratching arm
clapping hands (long) scratching back of the head
clapping hands (short) shaking head to say no (long)
coming closer to the webcam shaking head to say no (short)
crouching shoulder swipe
drinking something from a bottle shoulder warm-up
falling over shrugging (long)
feet apart shrugging (short)
fist bump (hold) sitting down
fist bump (preparation and hold) sitting on a chair
first bump (quick) small kicks while waiting
fixing hair (long, both hands) standing up
fixing hair (long, one hand) step feet together
fixing hair (short, one hand) stretching arms
give up gesture thumb down (hold)
going down on knees thumb down (preparation and hold)
grabbing a bottle (bottle visible from the start) thumb down (quick)
grabbing a towel (towel visible from the start) thumb up (hold)
grabbing an off-screen bottle thumb up (preparation and hold)
grabbing an off-screen towel thumb up (quick)
high five (hold) using towel to remove sweat
high five (preparation and hold) walking towards the webcam
high five (quick) waving (hold)
jump feet together waving (preparation and hold)
keeping hands in pockets waving (quick)
leaving plank position wiping face sweat on shirt
lying down after push-up yawning (covering mouth with hand)
lying down in random position yawning (long)
neck warm-up (with hands) yawning (short)

Dataset Citation Instructions


The dataset is intended for research purposes only. Please cite our paper if you use this dataset in your research:

Panchal, S., Bhattacharyya, A., Berger, G., Mercier, A., Böhm, C., Dietrichkeit, F., Pourreza, R., Li, X., Maden, P., Lee, M., Todorovich, M., Bax, I., Memisevic, R. (2024) Live Fitness Coaching as a Testbed for Situated Interaction

Dataset license

 

This dataset is intended for research purposes only and to support and contribute to the graph research community. The quality of the configuration space design and the collected execution times may be suboptimal and should not be considered as reference performances of the target device but rather as representative of the problem at hand for research purposes.


  

Data License Agreement - Research Use

Qualcomm AI Research

 

At Qualcomm AI Research, we are advancing AI to make its core capabilities – perception, reasoning, and action – ubiquitous across devices. Our mission is to make breakthroughs in fundamental AI research and scale them across industries. By bringing together some of the best minds in the field, we’re pushing the boundaries of what’s possible and shaping the future of AI.

Find out more about Qualcomm AI Research.

For any questions or technical support, please contact us at [email protected]

 

Qualcomm AI Research is an initiative of Qualcomm Technologies, Inc.

 

Connect with our communities

Stay ahead of the curve

Receive the latest updates, exclusive offers, and valuable insights delivered through the Qualcomm newsletter straight to your inbox.

Stay ahead of the curve

Receive the latest updates, exclusive offers, and valuable insights delivered through the Qualcomm newsletter straight to your inbox.

Qualcomm relentlessly innovates to deliver intelligent computing everywhere, helping the world tackle some of its most important challenges. Our leading-edge AI, high performance, low-power computing, and unrivaled connectivity deliver proven solutions that transform major industries. At Qualcomm, we are engineering human progress.

Stay connected

Get the latest Qualcomm and industry information delivered to your inbox.

Subscribe
Manage your subscription

© Qualcomm Technologies, Inc. and/or its affiliated companies.

Snapdragon and Qualcomm branded products are products of Qualcomm Technologies, Inc. and/or its subsidiaries. Qualcomm patented technologies are licensed by Qualcomm Incorporated.

Note: Certain services and materials may require you to accept additional terms and conditions before accessing or using those items.

References to "Qualcomm" may mean Qualcomm Incorporated, or subsidiaries or business units within the Qualcomm corporate structure, as applicable.

Qualcomm Incorporated includes our licensing business, QTL, and the vast majority of our patent portfolio. Qualcomm Technologies, Inc., a subsidiary of Qualcomm Incorporated, operates, along with its subsidiaries, substantially all of our engineering, research and development functions, and substantially all of our products and services businesses, including our QCT semiconductor business.

Materials that are as of a specific date, including but not limited to press releases, presentations, blog posts and webcasts, may have been superseded by subsequent events or disclosures.

Nothing in these materials is an offer to sell or license any of the services or materials referenced herein.