Spent today getting stuff in order so that face.com's performance on the Comcast MURI dataset can be systematically characterized. Most of the time went into building an HDF5 database storing a bunch of metadata for each image, including the face.com information. The data for the database was stored across a bunch of directories and file formats, so it took me a few attempts to get it constructed. Now that the database exists, I'm writing a script to automatically generate detection and miss examples by condition, including marking up the example image.
Also spent maybe 80min on db-class.org.
No comments:
Post a Comment