Modulation-Based Imaging Workshop
One week in the French Alps for the X-ray, optics and visible-light communities that work with modulation: gratings, masks, speckle, structured light. Talks in the morning. The rest of the day, a table big enough for everyone.
Why this workshop
Our field already has good conferences, and they are built around polished, best results. They leave little room for the other half of research: the dataset that behaves oddly, the method that works on one set-up and not on the next, the question two groups have been asking without knowing it.
MoBI 2027 is built around that half. It takes its spirit from scientific retreats and from the LINX initiative (Inserm, Monash, University of Canterbury), which already works this way. The location is remote on purpose. People get time to think, and the group is small enough that you can talk to everyone.
Scope
Modulation-based imaging is not specific to X-rays. A grating in a synchrotron beam, a diffuser in front of a camera, a structured illumination in a microscope: each shapes a wave and reads the result back. Coherence, modulator design and signal retrieval come up in every community, usually in different words.
Monday's two tutorials, one per community, give everyone the vocabulary of the other side before the week starts in earnest.
Not to scale · scope still open to more communities
Programme
Monday afternoon opens with two tutorials, one per community. From Tuesday, each morning has its own session, going from the source, through the modulator, all the way to the detector and beyond. Afternoons are working sessions for everyone.
Tutorials: X-ray phase imaging, and imaging with visible light
What does the beam bring to the measurement?
How do we shape, and design, the modulation?
What does the extracted signal actually mean?
How far can modulation take us?
Morning sessions run 09:00 to 12:30 with a coffee break. Invited talks last 30 minutes, selected talks 20 minutes, questions included. Lunch and dinner are taken together at the on-site restaurant. On Wednesday the working session ends at 16:00 for a hike, followed by the gala dinner.
Speakers
Each morning session opens with invited talks. Selected talks from the submitted abstracts complete the programme.
Afternoons
Everyone has something that does not quite work yet: a measurement that drifts, a model that holds until it doesn't, a result nobody in the lab could explain. Here, that is good material.
On Tuesday each participant gets five minutes to put one of these on a shared board. Groups form around the questions that draw people in, and they are mixed on purpose. The person who has already met your difficulty is often working at another wavelength, under another name.
About twelve lightning talks of five minutes. Then an open discussion, with the board filling up.
Groups of X-ray and visible-light people. Each defines a question, the data or simulation it needs, and what it hopes to show on Friday.
Round two, with the morning's modulator discussions fresh. A short break is left to each group, then the hike.
Shared datasets, real and simulated, run through several methods with a common set of figures of merit.
Ten minutes per group, then a roadmap: shared resources, benchmarks and next steps.
A first list from the scientific committee, to seed the week. It grows with the open problems you send through the call for abstracts. Yours may not be on it yet.
Comparing methods needs common inputs with a known answer. A sample described by its projected thickness and the structure of its scatterers can be simulated at any wavelength; real acquisitions of well-characterised phantoms would complete it.
Retrieval methods are implemented many times, in different languages and with different conventions. A shared open-source library with a common data model would help both communities.
Sensitivity, resolution, noise, bias and computation time are reported differently from one paper to the next. A handful of metrics, measured on the same data, would turn shared data and code into a benchmark.
Spatial and temporal coherence are described with different conventions in X-ray and in optics, and are rarely reported next to the results.
Many forward models assume full coherence, while real beams are only partially coherent. Where does the coherent model stop being valid, and what does a partially coherent model cost?
Polychromatic X-ray beams and white-light sources both blur the modulation and can bias the retrieved signals. Which corrections transfer from one field to the other?
Under visible illumination, objects are often pure phase objects; under X-ray illumination there is always some attenuation. Approximations valid in the visible cannot always be made with X-rays.
Gratings, checkerboards, random diffusers and speckle, coded or structured patterns: each is defended on its own terms. A fair comparison needs common criteria.
Absorbing modulators throw photons away; phase modulators and scatterers keep them but are harder to model. Dose for X-ray, light budget for the visible: the same trade-off.
Real modulators deviate from their design in profile, period and stability. How accurately must we know them, and how do we measure them in practice?
Simulation is used in both communities, but each validates it against different experiments. A common validation case would show which simulators can be trusted for design.
Integrating phase gradients is sensitive to noise, boundary conditions and low spatial frequencies. Reconstructing a wavefront from slopes is also a classical problem in optics.
Noise enters the forward model, the retrieval and the regularisation. How do we get useful uncertainty estimates without heavy Monte Carlo runs?
The signal depends on the set-up (propagation distance, modulation scale, wavelength) as well as on the sample. Which sample-intrinsic quantity can we extract?
Modulator, detector response, geometry, wavelength: which of these can be recovered from the data itself, and which need a dedicated measurement?
Energy-resolved detectors and hyperspectral cameras can separate absorption, phase, scattering and materials. What is gained over a single band, and at what cost?
Modulation can encode information below the pixel size. What sets the limit, and when is a claimed gain in resolution real?
Dark-field and scattering can depend on direction and scale. How do we describe them compactly, and how do band-resolved descriptions compare with tensor-like ones?
Polarisation-sensitive contrast in the visible may have equivalents elsewhere, and EUV, terahertz or neutrons could use the same ideas. What transfers, and what must change?
Networks trained on simulations or limited data can look convincing and still be biased. How do we validate them, and what should be reported with the results?
A balance between noise, the information we can still extract, and grating optimisation.
Bring the one that keeps you up. Someone here has probably seen it from the other side.
Place
The Centre Paul-Langevin (CAES du CNRS) is a recently refurbished centre at about 1,500 m, in a south-facing Savoie village near the Italian border. Rooms, restaurant, bar and meeting spaces are all on site, with a 200-seat auditorium and many breakout rooms, so the day does not scatter when a session ends.
In early June the valley is green, and the trails of the Vanoise National Park start at the door.
Photos: CAES du CNRS, caes.cnrs.fr. Restaurant photo: Alban Pernet (albanpernet.com).
Key dates
Abstracts and open problems are both sent through the workshop’s Sciencesconf site, using the abstract template: one page of text, up to two figures, and an optional page to describe an open problem.
Selected talks of 15 minutes plus questions, in the morning sessions.
Deadline 15 February 2027Submit an abstractAbstract template (.docx)The question, failure or dataset you would like help with. It goes on the shared board before the workshop and can become a Tuesday pitch.
Send an open problemAbstract template (.docx)More details soon: fees, travel support and how to register will be announced here.
Deadline 26 April 2027Scientific committee
Organised within the IRP LINX project.
