I am a Postdoctoral Research Scientist at Torr Vision Lab at University of Oxford. The focus to my research is to develop smart descriptor for Computer Vision application particularly Segmentation.
Our main contribution is construction of optimum learned descriptors for vision applications particularly segmentation. Traditional hand-crafted descriptors are not sufficiently invariant for vision tasks. Learned descriptors (from Neural Networks), on the other hand, cannot be tailored to region of interest and their receptive field is always rectangular. We have introduced learned shape-tailored local descriptors which have the invariance of learned descriptors (from Neural Networks) and their receptive field is totally contained and tailored to region of interest. Our descriptors achieve state-of-the-art performance on several image segmentation datasets.
Traditional local descriptors aggregate data in a fixed neighborhood around a pixel in an image which might aggregate data from different regions. We introduced a new descriptor, called Shape-Tailored Descriptors, which aggregates data only from the region of interest. This results in a significant improvement in the quality of the descriptor particularly near the boundary of textured regions. We came up with an algorithm where the descriptor and segmentation are jointly computed. Our algorithm for shape-tailored descriptors showed state-of-the-art performance in segmentation and tracking applications.
2015-2018 PhD Electrical Engineering, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia. CGPA 4/4.00
2012-2014 MS Electrical Engineering, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia. CGPA 4/4.00
2007-2011 BSc Electronic Engineering, GIK Institute of Engineering Sciences and Technology, Topi, Pakistan. CGPA 3.99/4.00
Email: naeemullah.khan@eng.ox.ac.uk