Vinay Ch ([email protected])Dr. Mehul Bhatt and Jakob Suchan at the Cognitive Vision Lab, University of Bremen
This project contains a pipeline to train, test and visualise fine grained activity recognition in videos. MPII Cooking Activities dataset contains 65 cooking activities in 44 videos continously recorded in a realistic setting and is publicly available. Activities are distinguished by fine grained body motions which have high intra-class variance.
This project uses openpose to get the pose tracks and is based on Sijie Yan, Yuanjun Xiong, and Dahua Lin's 2018 paper Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition. This work is part of a bigger project which aims to use multimodal features to detect and characterize emotion in videos.