No description, website, or topics provided.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
caffe
LICENSE
README.md

README.md

SwiDeN

This project is a CAFFE implementation for our ACMMM 2016 paper describing Switching Deep Networks(SwiDeN), which is a novel`deep depictive style-based switching mechanism which appropriately addresses the depiction-specific and depiction-invariant aspects of the problem.

License

This code is released under the MIT License (Please refer to the LICENSE file for details).

Citation

Please cite SwiDeN in your publications if it helps your research:

@article{2016arXiv160708764K,
Author = {Kiran Sarvadevabhatla, Ravi and Surya, Shiv and Kruthiventi, Srinivas.~S and 
Babu R, Venkatesh},
    Title = {SwiDeN : Convolutional Neural Networks For Depiction Invariant Object Recognition},
    Journal = {ArXiv e-prints},
    eprint = {1607.08764},
    Keywords = {Computer Science - Computer Vision and Pattern Recognition},
    Year = {2016},
    Month = {july},
   }

Dependencies and Installation

  1. Code for SwiDeN is based on CAFFE. This code was tested on UBUNTU 14.04 on the folowing NVIDIA GPUs: NVIDIA TITAN X, NVIDIA K40, NVIDIA K20.

  2. To install this version of CAFFE used to realize SwiDeN, install all the dependencies for CAFFE and then execute the following:

    $ git clone https://github.com/val-iisc/swiden.git
    $ cd swiden/caffe/
    $ make all 
    $ make matcaffe #if you have a MATLAB installation and want to link CAFFE to MATLAB
    $ make pycaffe  #if you want to link CAFFE to python

Datasets

We provide a link to the page on which the Photo-Art-50 dataset is hosted. Please cite the papers of the authors of the dataset if you use it. The models for testing SwiDeN can be found here. The train-test splits are located in /swiden/caffe/data/swiden_split/ .

Q&A

Please send message to shiv.surya314@gmail.com if you have any query regarding the code.