CVPR Daily - Tuesday

Carlo’s picks of the day: Carlo Dal Mutto is CTO of Aquifi Inc , working on advanced solutions that combine 3D and AI for removing errors in manufacturing and logistics. Carlo olds a PhD from the University of Padova, Italy. He is author of two books, several papers and is the inventor of over 20 patents regarding depth sensing, 3D reconstruction and deep learning applied to 3D data. “KCNN is a deep-learning based approach for extremely efficient keypoint detection. KCNN is a framework able to effectively emulate different keypoint detectors (e.g., KAZE, SIFT) and the proposed CNN architecture is suited for efficient implementation on embedded systems such as FPGAs, resulting in less than 1 millisecond computation time. KCNN received yesterday the best paper award at the Embedded Vision Workshop. We are very proud of this award and we really believe that efficient embedded implementation of deep learning techniques constitutes a fundamental pillar for many real-world applications.” For today, Tuesday 19 2 Carlo’s Picks Tuesday Orals: 08:50 (O1-1B ) [A7] “ Finding Tiny Faces in the Wild With Generative Adversarial Network ” 08:50 (O1-1C) [C13] “ A Certifiably Globally Optimal Solution to the NonMinimal Relative Pose Problem ” 09:40 (O1-1A) [A4] “ Learning by Asking Questions ” 14:50 (O1-2A) [A6] “ Deep Layer Aggregation ” 14:50 (O1-2C) [E10] “ Learning to Find Good Correspondences ” 14:50 (O1-2B) [C12] “ SPLATNet: Sparse Lattice Networks for Point Cloud Processing ” 14:50 (O1-2C) [E13] “ OATM: Occlusion Aware Template Matching by Consensus Set Maximization ” Spotlights: 15:48 (S1-2C) [G2] “ Deep Adversarial Metric Learning ” 15:48 (S1-2C) [F9] “ Self-Supervised Feature Learning by Learning to Spot Artifacts ” Demo: Carlo invites us with Paolo Di Febbo to Hall C demo: 10:10-12:30 “ KCNN: Extremely- Efficient Hardware Keypoint Detection With a Compact Convolutional Neural Network ”