welcome to Ahmad Humayun's webpage.

I am a Computer Vision researcher working with Jim Rehg and Irfan Essa as a PhD student at the Computational Perception Lab., Georgia Tech.

What excites me is that we know of at least one exceptional vision system: the brain - which theoretically gives Artificial Intelligence researchers a system to mimic. Interestingly, some neuroscientists have shown how critical is motion when our visual cortex learns to recognize patterns [1]. This is why I work with video data. We humans have developed our perception from a continuous stream of frames - not from individual images, so why should the machines we make be any different? Over the coming years I would like to build algorithms which can perceive video through machine learning. As a secondary aim I would like my systems to be extensible to even comprehend single images.

Biography: I graduated with a Masters in CG, Vision & Imaging in late 2010 from UCL. Here, I researched with Gabriel Brostow (aka Gabe) on detecting regions of occlusion in consecutive video frames. I also did a brief stint at The University of Warwick with Nasir Rajpoot, developing registration and dimensionality reduction techniques for cancerous tissue examined under Toponome Imaging System. Previously, I was stationed at LUMS SSE where I worked with Sohaib Khan . In my 3 years stay, I collaborated with biologists at LUMS SSE and MRC NIMR in developing tracking techniques for fluorescence microscopy.

Over the years, I have mainly tackled the problem of occlusion in my work. In the overall scheme of things I believe time is the missing dimension in most vision research. I am currently intrigued by problems where learning can help reveal more information about video sequences.

Apart from Computer Vision, I have had interludes into Systems research - working on Google's MapReduce with Umar Saif.

Resume/CV   (last updated: January 2014)

Publications    [Google Scholar | MS Academic]
Projects
  • New
    "RIGOR: Reusing Inference in Graph Cuts for generating Object Regions", IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2014. [Webpage + Code], [PDF], [BibTeX]
  • New
    "Video Segmentation by Tracking Many Figure-Ground Segments", IEEE Conference on Computer Vision (ICCV), Dec. 2013. [Webpage + Code], [Dataset], [PDF], [BibTeX]
  • "Learning a Confidence Measure for Optical Flow", IEEE Pattern Analysis and Machine Intelligence, May 2013. [Webpage + Code], [PDF], [BibTeX]
  • "A Novel Paradigm for Mining Cell Phenotypes in Multi-Tag Bioimages using a Locality Preserving Nonlinear Embedding", International Conference on Neural Information Processing (ICONIP), Nov. 2012. [Preprint PDF], [BibTeX]
  • "RAMTaB: Robust Alignment of Multi-Tag Bioimages", PLoS ONE, Feb. 2012. [link], [BibTeX]
     
  • "A Framework for Molecular Co-Expression Pattern Analysis in Multi-Channel Toponome Fluorescence Images", Microscopy Image Analysis with Apps. in Biology (MIAAB), Sept. 2011. [PDF], [BibTeX]
  • "Towards Protein Network Analysis Using TIS Imaging and Exploratory Data Analysis", Workshop on Computational Systems Biology (WCSB), June 2011. [Preprint PDF], [BibTeX]
  • "Learning to Find Occlusion Regions", IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2011. [Webpage + Code], [PDF], [BibTeX]
  • "Myosin motors drive long-range alignment of actin filaments", Journal of Biological Chemistry, Feb. 2010. [link], [BibTeX]
    this journal is ranked 7th by Eigenfactor across all scientific proceedings
Email: Voice: