Challenge Description

The rapid development of Augmented Reality (AR) and Virtual Reality (VR) technology has led to high demand on precise finger movement-based controllers for gaming and other applications. Immersive Holoscopic 3D imaging technology has been explored a lot recently and some small size sensors are now available. It can produce a 2D image with 3D information in it from lens array inside. This challenge will provide the opportunity to the research community to design finger micro-gesture recognition system using their own 3D image processing, feature extraction and classification methods, on a new built Holoscopic 3D micro-finger gesture dataset.

The dataset contains image sequences of different conventional finger micro-gestures (i.e. Button, Dial, and Slider) from 40 sujects.

The two sub-challenges will be

• Holoscopic 3D image based micro-gesture recognition (~6.6G data).

• Holoscopic 3D video based micro-gesture recognition (~40G data).

Participants will be allowed to apply their own 3D image pre-processing, feature extraction and classification methods in both sub-challenges. Baseline result will be given there soon.

The dataset was split into three sets for this challenge: training (20 subjects), validation (10 subjects) and testing (10 subjects). Participants are free to use either one or both data. The labels of the testing set will not be disclosed to the participants. For both sub-challenges, participants will need to adhere to the partition of training, validation and testing sets. In their papers, they may report on results obtained on the training and validation sets, but only the results on the testing set will be taken into account for the overall challenge results.

The dataset will be provided to participants after registrations. The participants are expected to register as a team to receive the dataset from us. The link will be sent to the participants after they have registered and signed the user agreement.

The participants can use and define their own 3D image processing, feature extraction, and classification on the training and development partitions. Finally, the testing partition will be provided without labels and the participants will be allowed a maximum of 5 submission attempts. The submission results on the testing set will be provided by email. After the challenge results submission deadline, the participants will be required to submit a workshop paper to describe their methods.

Only the results with an accepted workshop paper will be considered for the winner. The top 2 teams will be required to provide their code (it doesn’t have to be the source code) for further confirmation of their results. The best team will be awarded a winner certificate.