New version of Benchmark

• The code points to the pre-computed evaluation recently made available.
• This new version implements the measures used in the PAMI submission and in the following paper:
  Jordi Pont-Tuset and Luc Van Gool
  Boosting Object Proposals: From Pascal to COCO
  International Conference on Computer Vision (ICCV) 2015
• The databases are handled using the gt_wrappers presented in the ICCV code.

Code compatible with COCO

This new release contains the code accompanying the extension of the CVPR paper in the arXiv release.

The major changes are:

• Experiments are now compatible with COCO. We've created a wrapper that makes it easy to evaluate results on COCO in parallel.
• Non-maximum suppression added to the box generation, so now the box proposals are more competitive.

Please note:

• We have released the pre-computed results for COCO in the webpage of our project.
• To handle the amount of files of COCO, this post might come in handy.
• If you need to recompile, you'll have to add C++11 compatibility, check this post

Initial release, as used in our CVPR 2014 paper

Initial release, as it's in the project webpage. It has some known bugs that are solved in the current versions.

Bug fix

• Reproducibility solved, now results are always exactly the same.
• The code does not crash when empty UCMs are generated.
• The default parameters are now set in the full package.