This work aims to provide a small pool of regions for an image, that are likely to contain every object in the image, regardless of category.
- 11/23/2010 - Precomputed proposals for Pascal VOC 2010 + Support for 64-bit Windows
- 9/21/2010 - Included support for 32-bit linux
- 9/20/2010 - The website is now up!
- The code to generate proposals for an image is now available.
- Category Independent Object Proposal Code PROP_code.tar.gz (15MB)
- The code to generate a ranked pool of candidate object regions for each image.
- This code is written in Matlab, with supporing compiled mex functions.
- I've tested it on 32 and 64 bit linux machines (Matlab 2009b i.e. 7.9)
- Annotated Dataset PROP_BSDS_ann.tar.gz (390KB)
- These are the annotations we added to the Berkeley Segmentation Dataset.
- These annotations provide the segmentations for each object, and respect the region boundaries of the original dataset.
- PASCAL VOC 2010 Precomputed Proposals pascal_2010_proposals.tar.gz (15.6GB!)
- Precomputed regions for each image in PASCAL VOC 2010 (train+val+test). Please note that this is a very large file.
-  “Category Independent Object Proposals” [pdf][slides]
- Ian Endres, and Derek Hoiem
- ECCV 2010.
This research is supported by NSF IIS-0904209: Physically Grounded Object Recognition.
Please contact iendres2 -at- uiuc.edu with any questions.