Citation For predicting linear classifiers and the text augumentation CUB and flower dataset, please cite [1] and [2]. 1- Mohamed Elhoseiny*, Babak Saleh, Ahmed Elgammal, “Write a Classifier: Zero Shot Learning Using Purely Textual Descriptions”, ICCV 2013. |
paper  , poster
| 
2- Mohamed Elhoseiny*, Babak Saleh, Ahmed Elgammal, “Heterogeneous Domain Adaptation: Learning Visual Classifiers from Textual Description”, VisDA Workshop, ICCV 2013.Code Further documentation an code after acceptance of the journal version.
short paper ,
Text Descriptions
The augmented text could be downloaded from 1202-Elhoseiny-sup.zip
Splits for comparison
CUB2010 (zero shot splits): SplitsLCUB2010.zip
There is a train.txt,validation.txt, and test.txt for each fold. If you don’t need to validate parameters you can use the union of train.txt and validation.txt for training. test.txt include only examples of unseen classes (the class is not in either train.txt or validation.txt). Text and Visual Features
Text and Visual features for CU Birds and Oxford datasets could be downloaded here ICCV13 features
Code
ICCV methods code.zip (7k)
This is code for each baseline method separately. I will try to provide an example for how to use them later here. The purpose of this code is to give
an idea how the method works. The optimization is based on IBM CPLEX package, it needs to be installed. It is free for academic use.
The DA method we implemented using gradient descend is included. Also the quadratic program in included. The
TGP baseline could be reproduced by the code here http://www.maths.lth.se/matematiklth/personal/sminchis/code/TGP.html.
Talks
1-"Zero Shot Learning for Visual Object Category Recognition: Cutting Edge Approaches", at Faculty of Computer and Information Sciences, Ain Shams University, Jan 4th 2014. (presentation  )
Write a Kernel Classifier
We developed our work to be generalized and support kernels in both visual and unstructured text domains. We also proposed a distributional semantic kernel in the text domain. The details of the generalized kernelized framework that allow a generic kernel for both visual and text domain, and for a proposed distributional semantic kernel for unstructured text descriptions, please visit this link .
|
|
Mohammad Elhoseiny, Nov 27, 2013, 8:46 PM v.2 Mohammad Elhoseiny, Sep 20, 2014, 8:46 PM v.2 |
