Challenges

Abstract

We present the HANDS22 Challenge, a public competition designed for the evaluation of the tasks of 3D hand and object pose estimation during interaction. Hand-hand and hand-object are the most common forms of interactions using human hands and understanding the 3D poses of hands and objects is critical to building AR/VR systems or robotic applications. We introduce two separate tasks for these scenarios and evaluate the performance of the participating methods. Winners of the tasks and prizes will be announced and awarded during the workshop and results will be disseminated in a subsequent challenge publication.

Our first task focuses on hand-held object pose estimation. While several methods have been developed for estimating object poses, very few works have focused on hand-held object pose estimation using the hand context. To encourage further research in this direction, we introduce Task 1: Hand-held Object Pose Estimation using HO-3D dataset, which contains 3D pose annotations of hand and object. In this task, the participants are required to estimate the pose of the hand-held object from an RGB image and are encouraged to utilize the provided (both train and test set) ground-truth 3D poses of the hand.

Our second task is aimed at two-hands pose estimation during hand-hand and hand-object interactions. Many hand pose estimation methods rely on 3D pose annotations for training which are difficult to obtain especially during interactions, thus limiting their application in real-world scenarios. To address this shortcoming we introduce, Task 2: Semi/Self-supervised Two-hands 3D Pose Estimation during Hand-object and Hand-hand interactions using the recently introduced Assembly101 dataset. The dataset provides mutli-view captured videos of hand-object interaction during assembling and disassembling of toy-vehicles without pose annotations. The accuracy of the participating methods is evaluated on a manually-annotated test set.

Winners and prizes will be announced and awarded during the workshop.

Please visit the challenge pages for more details.

Challenges Page

• Task 1: Hand-held Object Pose Estimation

• Task 2: Semi/Self-supervised Two-hands 3D Pose Estimation

Participation

We follow the rules of previous challenges and more details can be found in the challenge page.

• Submission deadline: Oct. 10. 2022 Oct. 15. 2022

• To participate and download the dataset please fill the form (closed) and accept the terms and conditions.

• Submissions are processed through CodaLab (register using the same email id provided in the above form):

  • Task1

  • Task2

• Each team must register under only one user id/email id. Teams found to be registered under multiple IDs will be disqualified.

• In order for participants to be eligible for competition prizes and be included in the official rankings (to be presented during the workshop and subsequent publications), information about their submission must be provided to organizers. Information may include, but not limited to, details on their method, synthetic and real data use, architecture and training details. Check previous challenge publication to have an idea of the information needed.

• Winning methods may be asked to provide their source code to reproduce their results under strict confidentiality rules if requested by organizers/other participants.

• For each submission, participants must keep the parameters of their method constant across all testing data for a given task.