metadata

size_categories:
  - 10K<n<100K
task_categories:
  - image-to-image
dataset_info:
  features:
    - name: lensless
      dtype: image
    - name: lensed
      dtype: image
  splits:
    - name: train
      num_bytes: 5600452730
      num_examples: 24000
    - name: test
      num_bytes: 230987060
      num_examples: 999
  download_size: 5873531153
  dataset_size: 5831439790
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
tags:
  - lensless
  - computational-imaging

For future training, it is recommended to use this normalized version of the dataset.

More accessible (6GB instead of 100GB) copy of: https://waller-lab.github.io/LenslessLearning/dataset.html

Original license: https://github.com/Waller-Lab/LenslessLearning/blob/master/LICENSE

This dataset was prepared with this script.

After cloning and installing LenslessPiCam, ADMM reconstruction can be applied to the dataset with this script (handles dataset downloading from Hugging Face).

python scripts/recon/dataset.py

The models in this collection use the original DiffuserCam MirFlickr dataset during training. This dataset tries to replicate that version of the dataset (using NPY files during training). One slight different is that we were required to subtract the mininum of value the numpy arrays so that they could be stored as viewable images.