Selective Search For RCNN

Matlab Scripts and functions to generate selective search boxes for images used in RCNN based methods.
Support Windows and Linux.
Support parallel speedup.

Quick Demo

Run demo.m in Matlab.

Generate boxes for your dataset

1. Modify idx_file, format_string in start.m
   Such as for VOC2012 validation dataset:

   idx_file = 'path/to/VOCdevkit/VOC2012/ImageSets/Main/val.txt';
   format_string = 'path/to/VOCdevkit/VOC2012/JPEGImages/%s.jpg';
   

2. Run start.m, and results will be saved to output.mat

Boost with parallel computing

1. Install Parallel Computing Toolbox
2. Enable Automatically create a parallel pool in Matlab preferences.
3. Enjoy huge improvements!!
   Here is a Example for 20 images:
   

Disclaimer

1. selective_search_boxes.m is copied from rbgirshick/rcnn under BSD-2-Clause License.
2. SelectiveSearchCodeIJCV is downloaded and modified from SelectiveSearchCodeIJCV.zip. See LICENSE.txt for more details.