Current Results

FaceScrub | FGNet

Get Started

• Fill out this form to gain access to dataset. You will receive access information soon after if approved.

• View readme.txt included in development kit. It explains how to test your algorithm after training.

• Please read the FAQ below.

Experiment

• Identification and Verification

  1. Download our training set containing 672K identities and train your algorithm on the data (link)

  2. Download MegaFace disjoint distractors (link), FaceScrub (.tgz), FGNet (.tar.gz), and development kit (.zip) (.tar.gz)

  3. Run your algorithm (trained on our data) to produce euclidean features for all datasets. The development kit includes a script to save to the correct feature format.

  4. Run our experiment evaluation script (included in the development kit) with 10, 100, 1000, 10000, 100000, 1000000 distractors. Updated instructions for Challenge 2 evaluation can be found in the development kit README.

  5. Upload results into the google drive folder you received with access information. Please also upload links to features files for the full FaceScrub, FGNet, and disjoint MegaFace datasets

  6. Complete the submission form.

  More information about the experiment and the development kit files can be found in the development kit readme

Necessary Files

• Datasets

  • Our training dataset (link)
  • Our disjoint distractor dataset, which contains 1M distractor images that are disjoint from MF2. (link)
  • FaceScrub dataset (by downloading this dataset you agree to the conditions of the creators of the data)
    • FaceScrub (.tgz) — 16.4 GB
  • FGNet dataset (by downloading this dataset you agree to the conditions of the creators of the data)
    • FGNet (.tar.gz) — 45 MB

• Training Set

  • You must train with our training set. (link)
  • Our goal is to explore which algorithms can utilize the same large scale training data and perform the best.

• Linux Development Kit (.zip) (.tar.gz) — 16 MB

  To test your resultant algorithm.

• Required Others

  • OpenCV (link)

    Open source computer vision and machine learning software library

Timeline

• Challenge release: Aug 30, 2016, 6PM Pacific Time

• Submissions due: Sep 30, 2016, 6PM Pacific Time

• Results published online: Oct 15, 2016

  Feel free to use the results in your CVPR submission

Frequently Asked Questions

• Why are there fewer FaceScrub feature files in your features than in the whole set?

  • We use a subset of FaceScrub for initial tests (to speed up the testing) but will use the full FaceScrub for additional tests, please compute features for the full set.

• Do we need to submit our features for the full Megaface Disjoint set, FGNet, and Facescrub? Even if the DevKit probe lists don't use all of them?

  • Yes, please submit a link to all the megaface features, FGNet, and FaceScrub features.

Please cite the paper if you use our code, results, or dataset in a publication (link)