DDFF 12-Scene dataset consists of 720 lightfield images and coregistered depth maps.

  • Lightfield: 4D lightfield images; each of which has 9 × 9 × 383 × 552 undistorted subapertures Images are saved as numpy arrays and can be loaded as follows:
    import numpy as np
    lf = np.load('LF_0001.npy')
  • Lightfield-mat: 4D lightfield images in Matlab format
  • DepthRegistered: registered depth maps; recorded in meters and scaled by a factor of 1000. Depth images are saved in uint16 bits and only available for "train" set:
    import cv2
    from PIL import Image
    # in meters
    depth = cv2.imread('DEPTH_0001.png', cv2.IMREAD_ANYDEPTH) * 0.001
    depth = np.array(Image.open('DEPTH_0001.png'), dtype=np.float) * 0.001
  • RawImage: raw images consist of Lytro ILLUM RAW formatted images
  • LF CalibPattern: calibration pattern for the Lytro ILLUM camera
  • WhiteImages: white images required by the Lytro Desktop


All data in the DDFF 12-Scene benchmark is licensed under a Creative Commons 4.0 Attribution License (CC BY 4.0).


[24-04-2018] – Trainval/Test hdf5 files added
[07-12-2017] – Lighfield images in Matlab format
[05-12-2017] – Lighfield calibration pattern and white images
[15-09-2017] – Lightfield images, registered depth maps and raw Lytro ILLUM images


  title     = {Deep Depth From Focus},
  author    = {C. Hazirbas and L. Leal-Taixé and D. Cremers},
  booktitle = {Arxiv preprint arXiv:1704.01085},
  month     = {April},
  year      = {2017},

Share on

Share: Twitter Facebook Google+ LinkedIn