4.0 KiB

Raw Blame History

Fooling Code

This is the code base used to reproduce the "fooling" images in the paper: Nguyen A, Yosinski J, Clune J. "Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images". In Computer Vision and Pattern Recognition (CVPR '15), IEEE, 2015.

If you use this software in an academic article, please cite:

@inproceedings{nguyen2015deep,
  title={Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images},
  author={Nguyen, Anh and Yosinski, Jason and Clune, Jeff},
  booktitle={Computer Vision and Pattern Recognition (CVPR), 2015 IEEE Conference on},
  year={2015},
  organization={IEEE}
}

For more information regarding the paper, please visit www.evolvingai.org/fooling

Requirements

This is an installation process that requires two main software packages (included in this package):

Caffe: http://caffe.berkeleyvision.org

Our libraries installed to work with Caffe
- Cuda 6.0
- Boost 1.52
- g++ 4.6
Use the provided scripts to download the correct version of Caffe for your experiments.
- ./download_caffe_evolutionary_algorithm.sh Caffe version for EA experiments
- ./download_caffe_gradient_ascent.sh Caffe version for gradient ascent experiments

Sferes: https://github.com/jbmouret/sferes2

Our libraries installed to work with Sferes
- OpenCV 2.4.10
- Boost 1.52
- g++ 4.9 (a C++ compiler compatible with C++11 standard)
Use the provided script ./download_sferes.sh to download the correct version of Sferes.

Note: These are patched versions of the two frameworks with our additional work necessary to produce the images as in the paper. They are not the same as their master branches.

Installation

Please see the Installation_Guide for more details.

Usage

An MNIST experiment (Fig. 4, 5 in the paper) can be run directly on a local machine (4-core) within a reasonable amount of time (around ~5 minutes or less for 200 generations).
An ImageNet experiment needs to be run on a cluster environment. It took us ~4 days x 128 cores to run 5000 generations and produce 1000 images (Fig. 8 in the paper).
How to configure an experiment to test the evolutionary framework quickly
To reproduce the gradient ascent fooling images (Figures 13, S3, S4, S5, S6, and S7 from the paper), see the documentation in the caffe/ascent directory. You'll need to download the correct Caffe version for this experiment using ./download_caffe_gradient_ascent.sh script.

Troubleshooting

If Sferes (Waf) can't find your CUDA and Caffe dynamic libraries

Add obj.libpath to the wscript for exp/images to find libcudart and libcaffe or you can use LD_LIBRARY_PATH (for Linux).

Is there a way to monitor the progress of the experiments?

There is a flag for printing out results (fitness + images) every N generations. You can adjust the dump_period setting here.

Where do I get the pre-trained Caffe models?

For AlexNet, please download on Caffe's Model Zoo. For LeNet, you can grab it here.

How do I run the experiments on my local machine without MPI?

You can enable MPI or non-MPI mode by commenting/uncommenting a line here. It can be simple eval::Eval (single-core), eval::Mpi (distributed for clusters).

4.0 KiB Raw Blame History