@PerCom 2023 WS BiRD
We are hosting an activity recognition competition, OpenPack Challenge 2022, using the OpenPack dataset at a PerCom 2023 Workshop! The task is very simple: Recognize 10 work operations from the OpenPack dataset.
High recognition accuracy is required to introduce action recognition technology into the industrial domain. However, in packaging operations, not only the combination of items to be packed change a worker's movements greatly but also a number of other difficulties affect the packing process, such as a wide variety of motion speed, body movements, and procedures which are highly dependent on the skill level of the worker. Develop the perfect activity recognition model using OpenPack's various modalities and contribute to advancing smart logistics centers!
Recognize 10 Work Operations(Temporal Semantic Segmentation)
- Cash Prize
- Travel Fee Support to attend Percom2023
Human Activity Recognition is beginning to be applied in industrial domains, targeting the improvement of manual labor efficiency. We know processes such as assembly and packaging inside factories or logistic centers still mainly depend on manual workers. To ensure a rapid response to changing demands, this trend is expected to continue. Therefore, quantifying manual labor is a crucial requirement to streamlining existing processes.
Workers inside logistic centers perform a series of sequential tasks to pack items. These working periods consist of actions like reading item labels or assembling the packaging boxes. Understanding details from each of these operations, i.e. its temporal duration or potential abnormalities, is crucial for optimizing the packaging process. However, period uniqueness related to item size or number of items included, make identifying such operation details a challenging task.
You’ll develop a model to recognize the operations that conform packaging work with data from 4 IMU streams, keypoint data, etc. The packaging work consists of 10 operations (i.e., activity classes) described below. To quantify the operations as precisely as possible, dense labeling is required. You must predict activity classes for each 1second-long time slot. You can use data from 11 subjects to develop your model. The test data consists of 6 subjects (4 distinct subjects).
If successful, your work will help the ubiquitous research community improve current smart factories and better integrate human factors into the smart factory optimization process.
Task (Technical Details)
Plese visit the following page for more technical details such as task definition, evaluation procedures, and rules.
A. Cash Prize
The awards ceremony will be held at Percom 2023 WS (BiRD2023). We would like to hand out the cash prizes in this ceremony.
- 1st: 100,000 JPY
- 2nd: 40,000 JPY
- 3rd: 20,000 JPY
B. Travel Fee Support
At least one person from the top-3 top-5 team strongly encouraged to attend the award ceremony. We will support the travel fee for the attendance up to 1 member (up to 750,000JPY/person) from each winner team.
As a condition to being awarded a Prize, a top-5 winner must fulfill the following obligations. The detail instructions will be sent to top-5 winner after the final submission deadline.
- Submit your code so that we can check for cheating.
- Submit a short report paper that describes the award methodology.
Rules for OpenPack Challenge 2022
A. Registration Required
Please submit team registration form before you join this competition even if you a solo challenger. This is required to get prizes. There are no limitation for the team size. However, if you are creating multiple teams within the same lab, be careful not to violate the rules of PRIVATE SHARING.
When you want to merge your team with other team, please contact OpenPack Challenge Admin Team via email.
B. No Private Sharing Outside Teams
Privately sharing code or data outside of teams is not permitted. It’s OK to share code if made available to all participants on the forum.
C. External Dataset
You may use data other than the competition data to develop and test your submission. However, you will ensure the external data is publicly available to everyone without any cost.
Thank you to everyone who participated, we had 33 teams and 106 registered userson colab. We received a total of 817 submissions during the period. (As of January 19, 2023)
The top five teams in the competition are listed below. Congratulations! Please see this table (Google Sheet) here for detailed scores.
Awards Ceremony @Percom2023 WS BiRD
The awards ceremony was held on March 13, 2023 at Percom2023 WS BiRD. The top 5 winners gathered to discuss their solutions in a poster session. Congratulations again!
Registration and Submission (Closed)
Before submitting your prediction, registration is required. You need to register your information to both (1) CodaLab and (2) Google Form. Please follow the procedures below.
- Solo challengers are also welcom! But don't forget to register you as a solo team!
- No limitation to team size. (But be careful about violation of NO PRIVATE SHARING policy when you make multiple teams.)
Steps for Registration
- [Every Members] Make your codalab accounts. All members should have thier own accounts.
- [Every Members] Register to this challenge from Participate tab in the codalab page.
- [Team Leader] Submit the Google Form for team registration.
- (Wait for a while... Admin team will check and aprove your team.)
- You got email from codalab about approval and you can submit your prediction!