Deep Learning Methods for Vision

CVPR 2012 Tutorial

Speakers: Rob Fergus (NYU), Honglak Lee (Michigan), Marc'Aurelio Ranzato (Google) Ruslan Salakhutdinov (Toronto), Graham Taylor (Guelph), Kai Yu (Baidu)

Course Webpage

Tiny Images Dataset download

The entire 80 million Tiny Image dataset can be downloaded here.

Removing Camera Shake from a Single Image. Code for SIGGRAPH 2006 paper.

The code package includes the Matlab source code used in the paper, a comprehensive README file and sample images. Please fill in the form here and I will email you the code.

Fast 2D image convolution MEX file for Matlab under Linux.

Wrapper for Intel's Integrated Performance Primitives (IPP) libraries, also using OpenMP multi-threading. Requires 64-bit Intel CPU, up to date IPP libraries and Intel Compiler. Runs 5x-25x faster than Matlab's conv2 (or fft2 if kernel is large), check out the timings. MEX file source, MEX configuration, Matlab test script.

Learning Object Categories from Google's Image Search. Datasets from ICCV 2005 paper.

The .tgz file below contains the images of the seven categories collected from Google which were used in our ICCV'05 paper, "Learning Object Categories from Google's Image Search".

Google images and labels from ICCV'05 paper

The 7 classes are: Airplane, Cars Rear, Face, Guitar, Leopard, Motorbike, Wrist Watch. In each subdirectory, the labels are contained within a Matlab file, Ground_Truth.mat. This has a tri-state label for each image. 0 = Junk, 1 = Intermediate and 2 = Good Example. The labeling was performed by an individual who has no knowledge of the algorithm.

Google validation set from ICCV'05 paper

Train/test frame indices from ICCV'05 paper

Caltech test data: Guitars, Leopards, Wrist Watches used for testing in ICCV'05 Google paper

Constellation Model. Code for CVPR'03 and CVPR'05 papers.

I have code available for distribution for both the CVPR'03 paper "Object Class Recognition by Unsupervised Scale-Invariant Learning" and also the CVPR'05 paper "A Sparse Object Category Model for Efficient Learning and Exhaustive Recognition". The CVPR'03 code is Linux only. The CVPR'05 code works under both Windows and Linux and uses the VXL libraries. Please email me if you would like the code.

Oxford Robotics Reading Group on EM

Here are some materials from the reading group I gave on EM in Oxford on 24/05/05.

Powerpoint slides. Demo Matlab code for 2-D Gaussian mixtures