60 likes | 167 Vues
In Week 7, Alvaro Velasquez progressed in using KSVD to learn a dictionary for cuboids and obtained a sparse coefficient matrix through sparse coding with graph regularization. This framework supports the clustering of cuboids into 6, 8, and 20 clusters using k-means. Moving forward, the focus will be on enhancing the geometry of the image volume by optimally selecting initial cluster centroid locations through a max voting scheme applied to the sparse coefficient matrix. Work is also underway to apply this framework to sequences of temporal blocks, starting with a single block of 4 frames.
E N D
Progress Report: Week 7 Alvaro Velasquez
Accomplishments • Learned a dictionary for the cuboids using KSVD. • Obtained a sparse coefficient matrix using sparse coding with graph regularization (This is the same framework that was used for the tracking in previous presentations). • Clustered cuboids into 6, 8, and 20 clusters using kmeans.
Work for this Week • Further exploit the geometry of the image volume by choosing the initial cluster centroid locations using a max voting scheme on the sparse coefficient matrix. • Apply framework to a sequence of temporal blocks (We are currently using a single temporal block consisting of 4 frames).