Vectors of Locally Aggregated Tensors (VLAT)

   David Picard, Philippe-Henri Gosselin and Romain Negrel
   ETIS CNRS UMR 8051, France

Download datasets and related data

Before running demonstrations, you have to download data related to the corresponding dataset

Don't forget to download and compile the software !

For the Holidays dataset (linux commands):

Standard VLAT Demonstration

This demonstration will compute standard VLAT features, as proposed in [ICIP11].

To run the demo, execute the following command in the software directory:
  bash demo_sift.sh ~/Datasets/holidays_sift

Compact VLAT Demonstration

This demonstration will compute compact VLAT features, a method proposed in [ICIP12] to dramatically reduce the size of VLAT features. Note that compact VLAT can also be used for image categorization, as presented in [ICPR12].

To run the demo, execute the following command in the software directory:
  bash demo_sift_compact.sh ~/Datasets/holidays_sift

Wise VLAT Demonstration

Wise VLAT features is an improvement of standard VLAT using a cluster-wise processing of visual space. More details can be found in [MM13].

To run the demo, execute the following command in the software directory:
  bash demo_sift_wise.sh ~/Datasets/holidays_sift

Wise VLAT can be compacted as well:
  bash demo_sift_wise_compact.sh ~/Datasets/holidays_sift

Packed VLAT Demonstration

This demonstration will compute packed VLAT features, as proposed in [CVIU13]. Packed VLAT also aims at reducing the size of VLAT features. They lead to larger features than Compact VLATs, however they can pack any VLAT, without the need of projector or similar data. Note that they can also be used to speed up the computation of Compact VLATs.

To run the demo, execute the following command in the software directory:
  bash demo_sift_packed.sh ~/Datasets/holidays_sift

XHTML 1.1