AVCGR – Audio Visual Conversational Gesture Recognition
The aim of the task is to recognize the interlocutor’s gestures in conversational interactions. The data has been collected within the NOMCO project [1], and concerns first encounter dialogues in Finnish and Swedish languages.
Evaluation metric
To evaluate the results, a frame-based counting will be applied. The recognition method should output one of the annotated classes or “no class” for each frame. This will be compared to the ground truth in order to build a confusion matrix. Thus, the evaluation is similar to the AVRGR task.
[1] The NOMCO project. http://sskkii.gu.se/nomco/.