Update Compression for Deep Neural Networks on the Edge

An increasing number of artificial intelligence (AI) applications involve the execution of deep neural networks (DNNs) on edge devices. Many practical reasons motivate the need to update the DNN model on the edge device post-deployment, such as …

Occlusion-Robust Object Pose Estimation with Holistic Representation

We propose novel techniques to make object pose estimation remarkably robust against occlusions.

Physical Adversarial Attacks on an Aerial Imagery Object Detector

Deep neural networks (DNNs) have become essential for processing the vast amounts of aerial imagery collected using earth-observing satellite platforms. However, DNNs are vulnerable towards adversarial examples, and it is expected that this weakness …

End-to-End Learnable Geometric Vision by Backpropagating PnP Optimization

We propose BPnP, a network module for backpropagation through a PnP optimizer, as if the optimizer were a differentiable function.

Anomaly Detection via Neighbourhood Contrast

Relative scores such as Local Outlying Factor and mass ratio have been shown to be better scores than global scores in detecting anomalies. While this is true, our analysis reveals for the first time that these relative scores have a key shortcoming: …

Satellite Pose Estimation with Deep Landmark Regression and Nonlinear Pose Refinement

We proposed an approach to estimate the 6DOF pose of a satellite. Our method won the first place in the Kelvins Pose Estimation Challenge organised by the European Space Agency (ESA).

Neighbourhood Contrast: A Better Means to Detect Clusters Than Density

Most density-based clustering algorithms suffer from large density variations among clusters. This paper proposes a new measure called Neighbourhood Contrast (NC) as a better alternative to density in detecting clusters. The proposed NC admits all …