DeltaVision
DeltaVision widefield deconvolution microscopes and OMX super-resolution systems (Applied Precision, later GE Healthcare and Leica) acquire to .r3d files, and their softWoRx software writes deconvolved (_D3D.dv), projected (_PRJ.dv), volume-rendered and FRAP results as .dv. Both are the same format: an MRC-like header followed by the image sections.
Derived from hex dumps of corpus files and the documentation of the mrc Python package (BSD-3-Clause, Sebastian Haase and Talley Lambert). mrc and Bio-Formats (GPL) are run only as black-box oracles (see Validation). Every name below is ours. See docs/provenance/deltavision.md. Format id deltavision, crate openreadout-dv.
What OpenReadout returns
One image: X × Y = the section size, C = the wavelengths, Z = sections ÷ (wavelengths × time points), T = the time points. Planes are returned in stored row order with the samples converted to little-endian. dimension_order gives the file’s section order (XYZTC, XYCZT or XYZCT). The physical pixel size is dx, dy (and dz when there is more than one Z plane) in µm. Each channel’s name is its wavelength from the header; emission_nm and excitation_nm come from the extended header of the channel’s first section (the header wavelength when the extended header has none), exposure_ms from its exposure time. time_increment_s is the median step between the time stamps of consecutive time points (first wavelength, first Z); stamps that are not plausible seconds are left out (the FRAP series stores −1.05 × 10⁹ in its first section).
extra: section_order (ZTW, WZT, ZWT), pixel_mode, byte_order, lens_id (the objective’s id number), image_type, wavelengths_nm (as written; FRAP series use negative values for their photobleaching channels), titles (the processing history softWoRx writes: IMGCORR, DECON3D, X:(1 512) Y:(1 512) Z:(1 40 1)), written_by (softWoRx deconvolution when a title line starts with DECON3D, softWoRx image correction when one starts with IMGCORR, acquisition otherwise), extended_header {integers, floats per section}.
Header (header.rs, 1024 bytes)
The byte order comes from the identifier −16224 at byte 96 (a0 c0 little-endian, c0 a0 big-endian).
| offset | type | our name | meaning |
|---|---|---|---|
| 0, 4, 8 | i32 | width, height, sections |
section size and count |
| 12 | i32 | mode |
0 uint8, 1 int16, 2 float32, 4 complex float32, 5 int16, 6 uint16, 7 int32; 3 (complex int16) and others are refused |
| 16–36 | i32 × 6 | start, sampling |
start indices and sampling |
| 40, 44, 48 | f32 | spacing_um |
pixel spacing x, y, z (µm) |
| 76, 80, 84 | f32 | intensity |
minimum, maximum, mean (first wavelength) |
| 92 | i32 | ext_len |
bytes of the extended header |
| 96 | i16 | (DV_ID) |
−16224 |
| 128, 130 | i16 | ext_ints, ext_floats |
integers and floats per section in the extended header |
| 136–156, 172, 176 | f32 | other_ranges |
minimum and maximum of wavelengths 2–5 |
| 160 | i16 | image_type |
0 or 100 normal, 1 tilt series, 3 averaged (projections), … |
| 162 | i16 | lens_id |
objective id |
| 180 | i16 | time_points |
|
| 182 | i16 | sequence |
section order: 0 Z, then T, then wavelength (ZTW); 1 wavelength, Z, T (WZT); 2 Z, wavelength, T (ZWT) |
| 184–192 | f32 × 3 | tilt |
tilt angles |
| 196 | i16 | wavelengths |
channels |
| 198–206 | i16 × 5 | wavelength_nm |
one per channel |
| 208, 212, 216 | f32 | origin_um |
z, x, y origin |
| 224 | 10 × 80 bytes | titles |
text lines |
The sections follow at 1024 + ext_len, each width × height samples.
Extended header (section_info)
One record per section, in file order: ext_ints 32-bit integers then ext_floats 32-bit floats (8 and 32 in every corpus file; the block is padded to a multiple of 1024 bytes). The first 14 floats (SectionInfo): photosensor reading, time stamp (s), stage x, y, z (µm), minimum, maximum and mean intensity, exposure time (s), neutral-density filter, excitation and emission wavelength (nm), intensity scaling, energy conversion factor. info --view full --vendor lists them for the first 4096 sections.
check
| code | severity | meaning |
|---|---|---|
truncated |
error | the file is shorter than header + extended header + sections |
trailing_bytes |
warning | the file is longer |
extended_header |
warning | the extended header is smaller than its records need |
A header without the identifier, with sizes ≤ 0, sections that are not a whole number of Z planes, or an unknown section order is refused at open (exit 4 or 6). A plane past the end of the file is a corrupt-file error (exit 4).
Validation (2026-10-19)
10 corpus files from 7 sources (OME sample images DV/siRNAi-HeLa, will/FRAP, elena/P-TRE, U2OS, alexia/cajal-bodies; Zenodo 5887370, 15919154): raw .r3d (big-endian) and processed .dv files (deconvolved, projected, volume renderings, FRAP); pixel modes 0, 1, 2 and 6; section orders 0, 1 and 2; 1 to 4 wavelengths, 1 to 31 time points, up to 150 sections. Every compared plane is bit-exact with mrc’s DVFile; Bio-Formats 8.5.0 reports the same sizes and pixel type for every file.
Vocabulary (every public identifier in openreadout-dv/src must appear here)
| identifier | meaning |
|---|---|
DvReader, DvDataset, FORMAT_ID, open, open_input, header, dataset |
reader, opened file (and its parsed header), format id deltavision, modules |
looks_like_dv, dv_endian, mode_type, parse |
detection, byte order, pixel mode, header parser |
DvHeader, endian, width, height, sections, mode, start, sampling, spacing_um, intensity, ext_len, ext_ints, ext_floats, image_type, lens_id, time_points, sequence, wavelengths, wavelength_nm, other_ranges, origin_um, tilt, titles |
the header (above) |
channel_count, time_count, z_count, section_of |
sizes and the section of a plane |
SectionOrder { Ztw, Wzt, Zwt }, from_code, name |
section orders |
SectionInfo, section_info, photosensor, time_s, stage_um, exposure_s, nd_filter, excitation_nm, emission_nm, intensity_scaling, energy_conversion, SECTION_FLOATS |
extended header |
HEADER_LEN, DV_ID, DV_ID_OFFSET, MAX_EXT, MAX_PLANE_BYTES |
sizes and limits |
written_by |
which program wrote the file, from the title lines |
How this reader was derived, file by file: provenance log.