Fast-RCNN-Geo

This folder contains an example implementation of Geo Model in Fast-RCNN [1] / MatConvNet. The Geo Model Fast-RCNN model that was trained on a large synthetic dataset with fractures and breakouts.

There are one entry-point scripts:

• fast_rcnn_demoGeo.m: run in MatConvNet.

Note that the code does ship with a proposal generation method using Selective Search [2].

The fast_rcnn_demoGeo.m code should run out of the box, downloading the model as needed.

***** Before *****

1. git clone MatConvNet version 1.0-beta25 on https://github.com/vlfeat/matconvnet
2. Installing and compiling the library on http://www.vlfeat.org/matconvnet/install/
3. git clone this repository in examples/
4. Extract all tar.gz
5. Download Geo Model Fast-RCNN on https://www.dropbox.com/s/dc4xnqs9ldnva9s/net-deployedGeo.mat?dl=0 or https://drive.google.com/open?id=1n3JNVhosoTefoN3rM_FHkUUqYOXBynaA

To demo code using the first GPU on your system, use something like:

run matlab/vl_setupnn
addpath examples/fast_rcnnGeo
fast_rcnn_demoGeo('gpu',1) ; % using GPU
fast_rcnn_demoGeo ; % using CPU

References

1. Fast R-CNN, R. Girshick, International Conference on Computer Vision (ICCV), 2015.
2. Selective Search Van de Sande, Koen EA, et al. "Segmentation as selective search for object recognition." ICCV. Vol. 1. No. 2. 2011.