Different types of motion data collections and benchmarks are vital for the development and assessment of motion processing techniques. In the course of our research, we have created several new resources for motion data processing, which we offer to the scientific community.
LSMB19: Dataset for benchmarking search and annotation
We propose a benchmark to evaluate search and annotation algorithms. The benchmark contains a motion dataset of 2 very long and continuous unsegmented 3D skeleton sequences, training and testing data for two modalities (cross-subject and cross view), 98 search queries and ground truth labels.
»Visit LSMB19 benchmark homepage
PKU-MMD-129 query set
The PKU-MMD-129 query set is a collection of 129 motion sequences from the PKU-MMD dataset, which were used for evaluating the effectiveness of similarity-based motion retrieval in our paper Efficient Indexing of 3D Human Motions. We encourage other researchers to use the same queries and compare their results to ours. The 129 query sequences were selected from the PKU-MMD as follows: for each of the 43 single-subject motion categories provided by PKU-MMD, we selected 3 examples covering the three available viewpoints – left, middle, and right. Within each category and viewpoint, the specific query sequences was selected at random. For each query, we evaluated k*NN search, where k*+1 is the size of the respective query category (disregarding the viewpoint). Result objects were deemed relevant when they belonged to the same category as the query (again, disregarding the viewpoint).
