Thursday, April 28, 2011

vision talks at CMU are the best

At CMU, we get some of the best people in object recognition to visit and give talks.  The CMU VASC seminar is a place where all the cool vision researchers come and advertise their own research.  Just last week Deva Ramanan gave a spectacular talk about his most recent part-based detector.  Deva made an everlasting impression on me as a first year PhD student -- 5 years ago he showed the world that visualizations of person's parts as marginals are much sexier than just boxes.

Deva's Sexy Part Marginals

Today, Rob Fergus gave a talk on his most recent deep learning research.  This research topic has been promulgated by hardcore machine learning titans such as Andrew Ng, Yann LeCun, and Geoff Hinton, so it will be exciting to see how Rob applies these ideas to object recognition.  Unsupervised Feature Learning seems pretty exciting; unfortunately, I just cannot take results on Caltech101/Caltech256 very seriously :-(

Deep Learning Learning Architectures


  1. Hi,

    I was wondering what was the presentation of Rob Fergus that gave to you in unsupervised feature learning. Is it published work, or not just yet?


  2. Hi Stratis,
    The most relevant paper (about deconvolutional networks) is the following:

    M. Zeiler, D. Krishnan, G. Taylor, R. Fergus. Deconvolutional Networks. CVPR 2010.


  3. What are the concerns about CalTech 101 and 256? 101 especially is a common benchmark in Machine Learning. What about Cifar 10?

  4. Caltech 101/256 represent a certain special but limited view of the world. In particular, these datasets generally present a single visual concept per image (highly stylized image) and it is unclear whether progress on these datasets will enable machines to understand the visual world in a way which rivals human performance.

    Further reading on dataset bias:
    A. Torralba, A.A. Efros, ``Unbiased Look at Dataset Bias'', Conference on Computer Vision and Pattern Recognition (CVPR), Colorado Springs, June 2011

  5. Here is a followup from this post:

    The paper which Rob Fergus talked about at CMU is the following:

    Adaptive Deconvolutional Networks for Mid and High Level Feature Learning
    Matthew D. Zeiler, Graham W. Taylor, and Rob Fergus
    International Conference on Computer Vision(November 6-13, 2011)

    Matthew Zeiler is the first author and at the time of the presentation, the paper wasn't out so I had given a citation to an earlier paper.