PhD Candidate, Hong Kong University of Science & Technology (HKUST)

What’s the point of studying differential equations? Can we not do away with them? I almost never see an application of those as a computer science or an ECE systems student. Is it like the analog systems, we study for the legacy reason. Can we not do away from differential equations. I used to side […]

*October 11, 2017*

Given a set of correct keypoint matches and a fundamental matrix, to optimize the coordinates of these key points such that they satisfy the epipolar constraint. A point (x,y) on the left image (pose: [I|0]) and (x’,y’) on the right image (pose: [R|t]). These points are undistorted and in normalized image coordinates. Having known the pose […]

*August 17, 2017*

[GitHub] Extracting keypoints from images, usually, corner points etc is usually the first step for geometric methods in computer vision. A typical workflow is: keypoints are extracted from images (SIFT, SURF, ORB etc.). At these keypoints descriptors are extracted (SURF, ORB etc). Usually a 32D vector at each keypoint. The nearest neighbor search is performed to […]

*April 29, 2017*

SLAM (Simultaneous Localization and Mapping) is one of the important practical areas in computer vision / robotics / image based modelling community. A SLAM system typically consists of a) odometry estimator (relative pose estimator), b) Bundle adjustment module, c) sensor fusion module (for visual-inertial system), d) mapping module. While there are several excellent resources, refer […]

*February 15, 2017*

Download PPT – Google Docs Vector of locally aggregated descriptors (VLAD) [1] is a simple and popular technique for computing a fingerprint of an image for place recognition. It basically forms say K=64 clusters of SIFT like descriptors (descriptors at SIFT feature points). Then, for every descriptor subtracts it from cluster center and adds it up. […]

*January 11, 2017*

RNN (Recurrent Neural nets) are used to model sequences. Unlike the usual feedforward nets which are stateless in terms on inputs, RNNs have memory. In particular, its inputs are the output of previous step and also new observation in current step. The basic RNN are notoriously hard to train. LSTM (Long short term memory) networks […]

January 18, 20180