Skip to content
GitLab
Projects Groups Snippets
  • /
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
    • Contribute to GitLab
  • Sign in
  • bob.bio.face bob.bio.face
  • Project information
    • Project information
    • Activity
    • Labels
    • Members
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
  • Issues 22
    • Issues 22
    • List
    • Boards
    • Service Desk
    • Milestones
  • Merge requests 4
    • Merge requests 4
  • CI/CD
    • CI/CD
    • Pipelines
    • Jobs
    • Schedules
  • Deployments
    • Deployments
    • Environments
    • Releases
  • Packages and registries
    • Packages and registries
    • Package Registry
    • Infrastructure Registry
  • Monitor
    • Monitor
    • Incidents
  • Analytics
    • Analytics
    • Value stream
    • CI/CD
    • Repository
  • Activity
  • Graph
  • Create a new issue
  • Jobs
  • Commits
  • Issue Boards
Collapse sidebar
  • bobbob
  • bob.bio.facebob.bio.face
  • Merge requests
  • !141

SCFace porting

  • Review changes

  • Download
  • Email patches
  • Plain diff
Merged Laurent COLBOIS requested to merge scface into master Aug 20, 2021
  • Overview 5
  • Commits 3
  • Pipelines 2
  • Changes 10

Hello, This introduces a Bob 9 port of the Surveillance Camera Face Dataset (SCFace). In particular, this ports 5 protocols from the previous bob.db.scface package.

Full listing and metadata

Additionally to the usual protocol folders, the uploaded scface.tar.gz also contains a listing.csv file that lists all images from the database (around 4000) with as much metadata as possible. For example, some images with pose variations in particular a present in the database but never used in the protocols, those are listed in this listing. Note that those have no landmarks annotations.

The metadata for each image contains:

  • Subject metadata : Age, Gender, Beard, Mustache, Glasses
  • Device metadata : Channel (IR or RGB), distance (close, medium or far for RGB cameras), capture type (mugshot or surveillance), camera number
  • Landmarks metadata : Eyes, Nose and Mouth annotations for most of the images
  • Pose metadata (L1, L2, L3, L4 for various left profiles, R1, R2, R3, R4 for various right profiles, frontal for frontal views)
Protocols

The available protocols are:

  • scface-close : Enrolling frontal RGB mugshot, probing with RGB surveillance cameras at close distance
  • scface-medium : Enrolling frontal RGB mugshot, probing with RGB surveillance cameras at medium distance
  • scface-far : Enrolling frontal RGB mugshot, probing with RGB surveillance cameras at far distance
  • scface : Enrolling frontal RGB mugshot, probing with all RGB surveillance cameras
  • scface-ir : Enrolling frontal IR mugshot, probing with IR surveillance cameras
Experiment

As a test, I ran evaluation of all protocols (dev set only) using the arcface model : scface-arcface.pdf

Todo list for myself:
  • Check correct "Channel" metadata for IR surveillance cameras
  • Write tests
Edited Aug 20, 2021 by Laurent COLBOIS
Assignee
Assign to
Reviewers
Request review from
Time tracking
Source branch: scface