rspcunningham/imaged-data

imaged Fourier ptychography data

Raw Fourier ptychographic microscopy (FPM) acquisitions from a low-cost Raspberry Pi microscope: a Pi Camera Module v2 (Sony IMX219) with its stock lens, illuminated by a 145-LED ring array. Each study is a full single-wavelength acquisition: one raw Bayer frame per LED, plus dark frames for most studies, with every capture's LED position and exposure recorded.

The data is laid out to be loaded directly by imaged, a PyTorch FPM reconstruction library, but the format is plain JSON and NumPy, so any FPM code can use it.

Studies

Study	Sample	λ (nm)	Mag / NA	Frame (W×H)	Captures	Notes
20240709-005435-usaf-g7	USAF target, to group 7	525	1.7 / 0.13	3296×2464	145 + 7 dark	float32, normalised per frame
20240709-005435-malaria-carolina-9	Malaria parasite slide	525	1.7 / 0.13	3296×2464	145	acquisition date unknown
20250812-164017-usaf-g9	USAF target, to group 9	525	2.81 / 0.18	2464×3296	145 + 7 dark
20250821-160234-usaf-g9-colour-red	USAF target, to group 9	625	2.84 / 0.18	2464×3296	145 + 7 dark	red LEDs of an RGB run
20250821-160234-usaf-g9-colour-green	USAF target, to group 9	525	2.84 / 0.18	2464×3296	145 + 7 dark	green LEDs of an RGB run
20250821-160234-usaf-g9-colour-blue	USAF target, to group 9	470	2.84 / 0.18	2464×3296	145 + 7 dark	blue LEDs of an RGB run
20260728-122052-usaf-g9	USAF target, to group 9	525	2.7 / 0.15	1232×1232	145 + 7 dark	sensor crop, analog gain 5
20260902-144551-usaf-g9-analog-gain-1	USAF target, to group 9	525	2.7 / 0.15	3280×2464	145 + 7 dark	analog gain 1
20261007-174403-usaf-g9-long-exposure	USAF target, to group 9	525	2.7 / 0.15	3280×2464	145 + 7 dark	4 s exposure throughout
20261007-174946-usaf-g9-30um-defocus	USAF target, to group 9	525	2.7 / 0.18	3280×2464	145 + 7 dark	sample defocused by 30 µm

Study names are <acquisition date-time>-<sample>[-<variant>]. usaf-g7 and usaf-g9 are two different USAF 1951 targets whose finest group is 7 or 9.

Layout

<study>/
├── info.json        # manifest: optics, Bayer layout, per-capture LED position and exposure
└── captures/*.npy   # one 2-D raw Bayer frame per capture, shape (height, width)

The full info.json schema is documented in INFO_JSON_SCHEMA.md. In short, all lengths are in metres and exposures in seconds. LED positions are relative to the sample centre, with +z pointing from the sample to the LED array. Dark captures have no LED position and match the (channel, exposure) pairs of the illuminated ones. Free-form study details (sample, camera, gain, notes) live in info.json → metadata.

Usage

Download one study (about 450 MB) and reconstruct it with imaged:

hf download rspcunningham/imaged-data --repo-type dataset \
  --include "20260728-122052-usaf-g9/*" --local-dir datasets

from imaged import PtychStudy

study = PtychStudy.load("datasets/20260728-122052-usaf-g9", dark_subtraction="average_all")
study.captures          # [illumination, H, W] demosaiced, dark-subtracted, exposure-normalised
study.illumination_kx   # per-capture illumination spatial frequencies

Known issues

• 20240709-005435-usaf-g7 frames are float32 in [0, 1] and appear to be normalised per image at acquisition, so relative brightness between LEDs is not preserved. Its captured_at timestamps are synthetic.
• 20240709-005435-malaria-carolina-9 has no recorded acquisition date; its timestamps were copied from the USAF study of the same era.
• The 2026 studies use the measured numerical aperture of 0.15 instead of the 0.18 the acquisition software recorded, except 20261007-174946-usaf-g9-30um-defocus, which still carries 0.18.

License

Released under CC0 1.0: no rights reserved. Attribution is appreciated but not required.