Skip to content

SYSUDataset data structure for easier clear filtering

Mathias Réus requested to merge data_struct into master

Add a convenient data structure SYSUDataset to use for importing data from the .mat file.

The SYSUDataset is composed of 3 lists : train_data, test_queries and test_galleries. These lists are filled according to a CUHK-SYSU adaptation. In short :

  • train_data is a list of TrainingID. In each TrainingID there are multiple frame detection of the same person ID.
  • test_queries is a list of TestQuery. In each TestQuery there is one frame detection of a person. The person should be searched in TestGallery detection frames
  • test_galleries is a list of TestGallery. In each TestGallery there is a list of detection frames. Each TestGallery has a corresponding TestQuery. This is the search space of the query.

Note: test_queries and test_galleries have the same length. zip(test_queries, test_galleries) can be iterate over to produce (test_query, test_gallery) pair. In this pair, the person inside the test_query must be retrieved in the test_gallery.

Additionally, there is the build_dataset notebook to run the detail of the code. Also, the SYSU_dataset_exploration notebook uses the code in the lib to exhibit the data structure.

Merge request reports