Roadmap · 24 October 2026
Where UDeep is going
UDeep is a draft standard, the UDeep Figure Profile, for figure files made from data. A file that follows it is an ordinary image that also records how the data became the picture. The code in the repository is its reference implementation. This page says what the project is for, the two tests it has to pass, and what comes next. What exists today is on the About page.
A plan, not a promise: nothing here has a date, and the order can change.
The aim
The ordinary way a figure says how it was made
HDR photographs reached everyone without anyone choosing a format: the standard came first, and then the phone's camera saved in it by default. UDeep aims for the same with figures made from data. The tools people already use would write it without being asked, viewers and journals would read and check it, and in time it would become a standard of the kind photography already has. The likeliest home is ISO's photography committee, as a third image state beside scene-referred and display-referred: data-referred. The standardisation page sets out that route.
Two tests
It has to be seamless, and it has to be worth it
Seamless
No extra step for anyone. The reader needs nothing: a UDeep figure is an ordinary JPEG or PNG and shows as the ordinary figure wherever HDR does not. Whoever makes the figure should not have to change how they work, whether they write code or not: most figures are made by people who never script.
Worth it
Measured on the 18 HDR figures on this site: the record of how a figure was made adds 1.5 to 1.7 KiB, under 1 % of the file; the range above white adds about half as many bytes again. The record is useful on every screen today. That the range helps people read figures has not been shown: a study is designed and not yet run (About).
What comes next
The roadmap
In order, without dates. Each step says what it gives.
-
Done
The standard and its tools
- The profile, draft v0.3what a figure file made from data must declare, with numbered checks and shared test vectors
- A reference implementationPython, a Rust core, Julia and the browser, all reproducing the same test vectors
- A validatorchecks a file against the profile, and reads a data record back as the numbers
- Checkable in a hundred yearsevery figure's link to its data taken with SHA-256 and SHA3-256 (profile §16)
- Where the declaration is storedin each format, written down (profile 8.8), with a file taken apart and conformance files that test it
-
Next
Seamless to make and to check
- A releaseon PyPI, with a DOI: installed in one line, and citable
- A figure without codeopen a TIFF, a FITS file, a 16-bit PNG or a table in the browser, choose where white sits, save the figure
- No changes to a scriptone line added to a matplotlib script, and its figures are saved as UDeep files
- Check a figure in the browserdrop a file on a page and see its graded check; later, checked as it loads
- Signingan optional C2PA signature with the author's own key, off unless asked for (Trust)
-
Later
Shown to work, and used widely
- The observer studywhether the range above white helps people read figures; it needs ethics approval and participants
- A papera software paper or a methods preprint
- More toolsFiji and ImageJ, napari, R, MATLAB and Plotly; then the software that comes with instruments
- Browsers that read the recordthe value under the pointer, the units, the caption and the check, shown by the browser itself
-
After the paper
A standard
- A proposalthrough BSI, the UK national body, to ISO's photography committee (ISO/TC 42), as the data-referred image state (route)
Parked: vector figures with HDR, one SVG with sharp axes and an HDR data layer, until browsers support HDR colours, which they do not yet.
Today
Where a UDeep figure can be seen
A UDeep figure is a standard HDR image with a little metadata, so it needs no support of its own to be seen. Wherever a screen and its software show HDR images, the range above white appears; everywhere else, the ordinary figure does, which is still the complete figure.
| Where | What shows |
|---|---|
| Chrome and Edge (version 137 or later), Safari 26 | the gain-map JPEG with its range above white, on an HDR screen |
| iPhone Photos | the gain-map JPEG as HDR (Annex B, B1) |
| Firefox | the ordinary figure: it did not show gain maps when checked in September 2026 |
| Any screen without HDR, and print | the ordinary figure, as it would have been made anyway |
| A standalone SVG loaded as an image | the gain map is dropped; an HDR AVIF inside it keeps its range (Annex B, B4) |
| A screenshot | the ordinary figure: a screenshot keeps no HDR |
| The record (mapping, units, caption, check) | read by the validator, the live view and the packages; no browser reads it natively yet |
On Windows, HDR has to be switched on in the display settings. The device checks show what your own screen does, and reports of what they show are welcome.
Further out
What may lie ahead
This section is speculation, not evidence: where the world may go, and what each would mean for UDeep.
- Every screen HDR, and brighter. The ordinary figure becomes the fallback. A UDeep figure's range grows with each screen's headroom with no change to the file, since each display applies the range up to what it can show.
- Machines read figures. Software that reads papers and charts guesses values from pixels today. A UDeep figure gives it the exact mapping, the uncertainty and the link to the data.
- A picture with no record looks suspect. Signatures say who made a file; UDeep's record answers whether a picture follows from its data, which signatures do not.
- Papers become live pages, and a figure becomes a view of its data. A UDeep data record, read back as the numbers, already treats a figure that way.
- Data in headsets, where brighter than white is native. The declared mapping does not depend on the display.
- Figures kept for a century. The profile already says what a check made a hundred years from now rests on (§16): no live service, no trust list and no single algorithm.
Beyond the paper
Where else it could be used
Wherever a picture is made from measurements, and either the data has more range than an ordinary picture holds, or people need to trust how the picture was made. A first list, not yet tested with anyone in these fields:
- Medical images outside the clinic: teaching, papers and patient portals, with soft tissue and bone in one picture and the window stated. Diagnosis needs regulated software, so this is communication, not diagnosis.
- Microscopy and blots: faint and saturated signal together, and an image that says how it was adjusted.
- Astronomy and space images, for papers and the public: bright cores without losing the faint parts.
- Weather, climate and Earth observation: storms, fires and heat maps, with the colour scale and the data source in the file.
- Thermal imaging and inspection: a report image that states its temperature or intensity scale.
- News graphics and public statistics: a chart or map that can show which data it came from. A check says that a picture follows from the data it names, never that the data is true.
- Cultural heritage: infrared and X-ray images of paintings and manuscripts, with their processing on record.
Comments on this roadmap are welcome in the Q&A discussions or the issue tracker.