Papers
2
Total Citations
10
H-Index
2
About
Ben Harwood’s research lies at the intersection of self-supervised learning, video representation, and 3D place recognition, with a focus on developing methods that exploit spatiotemporal structure without relying on costly human annotations. In his highly cited work on *Temporally Coherent Embeddings for Self-Supervised Video Representation Learning* (2021, 6 citations), Harwood introduced a novel framework that explicitly enforces temporal coherency in the embedding space, moving beyond indirect ranking or predictive losses to capture the inherent dynamics of unlabeled video. This contribution has been influential in advancing unsupervised video understanding. Complementing this, his paper on *Locus: LiDAR-based Place Recognition using Spatiotemporal Higher-Order Pooling* (2021, 4 citations) tackles the critical challenge of global localization in large-scale environments. By proposing a higher-order pooling mechanism for 3D LiDAR point clouds, Harwood’s method strengthens non-local constraints in SLAM systems, enhancing robustness in autonomous navigation. Together, these works demonstrate a consistent drive to leverage temporal and spatial cues for representation learning, earning him recognition as an emerging voice in computer vision and robotics.
Research Focus
Key Achievements
Top Papers
- 1
- 2