IMDb-Face, Megaface

IMDb‑Face Dataset

• Location: https://github.com/fwang91/IMDb-Face (IMDb-Face.csv)
• Run Command:
  • Download the IMDb-Face.csv file.
  • Execute python imdb_crawl.py, supporting multi‑process handling.
  • Parameters:
    • -c: whether to crop images.
    • -d: whether to delete the existing data directory.
  • If saving uncropped images, the corresponding bounding boxes will also be recorded in a bb.txt file.

Megaface Dataset

• Download:

  • Distractor and probe datasets.
  • Access link: http://megaface.cs.washington.edu/participate/challenge.html

• Structure:

  MEGAFACE -- distractors -- parent id -- ids -- images | |- json file for each image | |- facescrub -- ids -- images, bb.txt |- bb.txt

• Pre‑processing:

  • Apply face detection/alignment models.

• Generate bin Files:

  • Use gen_megaface.py with a trained face‑recognition model to generate bin files for distractor/facescrub images.
  • Parameters include paths for distractor and facescrub images, noise lists, output directories, checkpoint model, and file endings.

• Run Megaface Devkit:

  • Execute python run_experiment.py in terminal (requires at least 32 GB RAM).
  • Devkit must be downloaded from http://megaface.cs.washington.edu/participate/challenge.html.
  • Parameters include paths to distractor and probe feature files, file ending, and size settings.

• Notes:

  • Binary files (bin/Identification, bin/FuseResults) can only be run on OpenCV 2.4.