Neural style transfer, the family of algorithms that repaint one image in the visual language of another, has produced dazzling results on photographs and modern paintings. But Thangka poses a distinctive trap. These works follow rigid compositional rules: central deities, concentric halos, precise registers of landscape and sky. When a general-purpose algorithm restyles a Thangka, it tends to smear those contours, warp the geometry and sprinkle texture artifacts across the heterogeneous surfaces of silk, pigment and gold leaf. For an art form in which geometry itself carries religious significance, structural distortion is not a cosmetic flaw — it destroys the content.

DART attacks the problem with three integrated components. The first is Multi-Directional Strip Attention, or MDSA, an attention mechanism that models long-range directional dependencies. Strip attention slices the image into elongated bands and lets the network compare distant regions along specific orientations — horizontal, vertical and diagonal. That matters for Thangka because the defining structures are directional: the straight borders of a mandala, the radiating lines of a halo, the vertical axis of a seated figure. By attending along strips, the model can keep a contour coherent from one edge of the canvas to the other, rather than treating each pixel patch as an isolated fragment.

The second component, a Dynamic Kernel Feed-forward Network (DKFN), addresses the texture problem. Thangka surfaces are wildly heterogeneous: smooth areas of sky sit beside dense brocade patterns, fine gold outlines and stippled pigment. A conventional feed-forward layer applies the same operation everywhere, which flattens that diversity. DKFN instead adaptively aggregates identity features and multi-scale features, adjusting its processing to match the local character of the surface. In effect, the network learns to treat a smooth gradient differently from a brocade weave, preserving the crispness of each texture rather than blurring them into a generic middle ground.

The third piece is a Structure-Aware Multi-Scale Edge Loss, a training objective that explicitly rewards geometric fidelity. Loss functions define what a network optimizes for; by penalizing errors on edges at multiple scales, this loss pushes the model to preserve fine contours and details while it restyles the image. The result is a system that can change the painterly language of a Thangka — its palette, brushwork and surface treatment — while keeping the underlying drawing intact.

The team evaluated DART on a high-quality Thangka dataset and reported leading overall performance among the general-purpose methods they compared against. But the more striking results came from the stress tests. In cross-domain evaluation on MS-COCO, the standard benchmark of everyday photographs, DART transferred its capabilities without being retrained. A blind user study added subjective evidence: human evaluators, unaware of which system had produced which output, preferred DART’s results. And in qualitative tests on material the model had never seen — Dunhuang murals, the Buddhist cave paintings of Gansu, and European historical paintings — the framework generalized without fine-tuning, suggesting the architecture captures something fundamental about geometry-preserving restyling rather than memorizing the quirks of one dataset.

The implications reach beyond one art form. Digital preservation of heritage imagery is a growing field, and the tools available to it have mostly been borrowed from general-purpose computer vision. Those tools often fail quietly: a restyled mural looks plausible at first glance, but the iconography has drifted. A framework designed to preserve geometry could let conservators experiment with reconstruction and visualization while keeping the drawing — the part that carries the meaning — verifiably unchanged. The authors also frame DART as a practical approach for creative reuse, allowing contemporary artists to restyle heritage imagery for new works without distorting the source material’s structure.

Technically, DART is an end-to-end framework, meaning the whole pipeline from input image to restyled output is trained as one system rather than assembled from separately trained modules. That integration matters because the three components reinforce one another: the directional attention keeps structures coherent, the dynamic kernel network handles the texture diversity, and the edge loss keeps the training focused on geometric fidelity. The architecture is a reconstructive transformer — it rebuilds the image rather than merely filtering it — which gives it the capacity to synthesize new painterly detail while anchoring that detail to the preserved structure.

The work comes from the Key Laboratory of China’s Ethnic Languages and Information Technology of the Ministry of Education at Northwest Minzu University, with additional affiliation through the Gansu Provincial Engineering Research Center of Multi-modal Artificial Intelligence. It was supported in part by the National Natural Science Foundation of China under Grant No. 62366047 and by the Fundamental Research Funds for the Central Universities under Grant No. 31920240060. The funders had no role in study design, data collection, analysis or publication decisions, and the authors declare no competing interests. All materials are available from the corresponding author upon request, and the article is open access under a Creative Commons licence.

For a field that has spent a decade chasing photorealistic restyling, DART’s contribution is a change of priorities: fidelity to structure over spectacle. The system demonstrates that a transformer can be built to respect the compositional logic of an art form rather than impose a generic aesthetic on it. As heritage institutions digitize more of the world’s sacred and historical imagery, frameworks that preserve geometry while transforming style may become the standard tool — not just for Thangka, but for any art where the drawing is the doctrine.

Subject of Research: AI and Cultural Heritage

Article Title: DART: a dynamic all-directional reconstructive transformer for geometry-preserving Thangka style transfer

Article References: Zhang, X., Wang, N., Wang, Z., Yang, Y., Wang, Y., & Zhang, M. (2026). DART: a dynamic all-directional reconstructive transformer for geometry-preserving Thangka style transfer. npj Heritage Science. https://doi.org/10.1038/s40494-026-03007-3

Image Credits: AI Generated

DOI: 10.1038/s40494-026-03007-3

Keywords: Thangka, style transfer, transformer, cultural heritage, computer vision, DART, dynamic, all-directional, reconstructive, geometry-preserving, style, transfer

The ACM conference series occupies a unique position in planetary science. Unlike meetings devoted to a single mission or technique, it deliberately bridges communities that often work in parallel: dynamicists who model orbital evolution over millions of years, spectroscopists who infer composition from reflected sunlight, laboratory analysts who study meteorites in the clean room, and observers who track the fleeting trails of meteors burning in the upper atmosphere. The triennial cadence means each meeting functions as a census of the field, capturing which questions have matured into consensus and which remain open. The 2026 edition in Poznań continued that tradition, bringing together current and future generations of small body scientists, with a scientific programme assembled by an international Scientific Organizing Committee and local logistics managed by a committee based in the Wielkopolska region.

A recurring theme of the meeting was the flood of high-precision data now arriving from next-generation surveys and space telescopes. Work presented in the session literature includes studies of trans-Neptunian and Centaur populations informed by recent survey discoveries, analyses of the orbital structure of near-Earth asteroid populations, and characterizations of active asteroids and main-belt comets whose dust tails blur the classical boundary between asteroids and comets. The increasing sensitivity of ground-based facilities has pushed detections toward smaller and fainter objects, revealing populations that were statistically invisible a generation ago. For dynamicists, this means models of impact flux, delivery from the main belt and the Kuiper belt, and resonant transport can now be tested against observed size-frequency distributions rather than fitted to sparse catalogs.

Cometary science featured prominently, reflecting the maturation of remote sensing at unprecedented spatial resolution. Research discussed at the conference included observations of cometary activity captured with the James Webb Space Telescope, which has begun resolving the distributions of water, carbon dioxide and other volatiles within cometary comae, and studies of long-period comets making their first pass through the inner Solar System. These measurements bear directly on questions of where comets formed and how the primordial ice inventory of the outer disk was distributed during the giant planets’ migration. Compositional mapping of individual comets, combined with the in situ results from the Rosetta mission at comet 67P/Churyumov-Gerasimenko, is converging on a picture in which comets are heterogeneous aggregates assembled from material processed at different distances from the young Sun.

Planetary defence was another major strand of the programme, and one with particular urgency following the first deliberate alteration of an asteroid’s orbit. The Double Asteroid Redirection Test demonstrated that a kinetic impactor can measurably change the trajectory of a satellite asteroid, and follow-up work on the ejecta dynamics, momentum transfer efficiency and structural response of the target Dimorphos has reshaped how researchers model asteroid interiors. Presentations at ACM 2026 drew on this legacy as the European Space Agency’s Hera spacecraft travels toward the Didymos system to survey the impact outcome in detail. Related work on the physical properties of rubble-pile asteroids, their cohesion, porosity and rotational failure modes, informs both deflection planning and the broader understanding of how small bodies evolve through collisions and spin-up over Solar System history.

Meteor science, the third pillar of the conference’s remit, is undergoing its own quiet revolution driven by global camera networks and seismic monitoring. Modern fireball networks now triangulate meteoroid trajectories with sufficient precision to recover falls and compute pre-impact orbits, linking meteorites directly to their source regions in the main belt. Studies presented in the meeting literature include analyses of meteoroid fragmentation in the atmosphere, statistical treatments of fireball datasets, and the use of seismometers to detect and characterize bolide airbursts, a technique that extends monitoring to events too small or too remote for optical networks. Together these approaches constrain the flux of material reaching Earth and the physical structure of the parent bodies, connecting the meteorite record in laboratories to the asteroid populations observed telescopically.

Sample return missions have fundamentally changed the evidentiary basis of the field, and their results resonated through many sessions. Material returned from the carbonaceous asteroids Ryugu and Bennu has provided uncontaminated, space-weathered primitive material whose mineralogy, organic content and isotopic composition can be compared directly with meteorite classes. Work building on these samples, including studies of the volatile inventory and thermal history of primitive bodies, was among the research highlighted in the conference’s session literature. The comparison between returned samples and telescopic spectra of larger populations such as the D-type and P-type objects in the outer belt and Jupiter’s Trojan swarms remains a central challenge, and new infrared surveys of these populations are helping to close the gap between the few bodies visited by spacecraft and the hundreds of thousands catalogued from Earth.

The meeting also underscored the growing role of machine learning and large-scale computation in small body science. Automated classification of light curves, neural-network detection of moving objects in survey images, and large N-body simulations of asteroid family evolution are now standard tools, and several presentations examined how these methods perform on the data volumes expected from the Vera C. Rubin Observatory’s Legacy Survey of Space and Time. Orbit determination for potentially hazardous objects, impact probability computation and the propagation of uncertainties over decadal timescales all benefit from these advances, as do efforts to link meteorite falls to asteroid families through dynamical modeling. The methodological convergence across subfields, from comet photometry to meteoroid stream modeling, was one of the practical takeaways emphasized by attendees.

Beyond the science, the Poznań meeting made a deliberate effort to broaden participation. Support from the ESA Academy, the Europlanet Society, the NASA Planetary Data System Small Bodies Node, the UAE Space Agency, the Canadian Space Agency and industry partners enabled early-career researchers and scientists from underrepresented and underprivileged groups to attend, and the UAE Space Agency sponsored the conference’s Best Poster Award. Financial backing from Poland’s Ministry of Science and Higher Education through the Wektory Nauki programme, the Wielkopolska regional government and the Initiative of Excellence – Research University programme of Adam Mickiewicz University underwrote the meeting itself. Organizers noted that the resulting exchange helped foster new collaborations across the global small body community, a stated aim of a series that has historically rotated among host countries to keep the field international.

As the 15th edition closed, the field it surveyed looked markedly different from the one that gathered three years earlier. Spacecraft are en route to metallic and primitive asteroids, survey telescopes are cataloguing small bodies at rates that outpace manual analysis, and laboratory measurements of returned samples are rewriting textbooks on Solar System formation. The questions that will define the next ACM conference, expected in 2029, are already taking shape: how efficiently kinetic impactors transfer momentum to real rubble piles, how the isotopic diversity of meteorites maps onto asteroid spectral classes, and how the smallest members of the Solar System’s debris population record the giant planets’ early history. For the community that met in Poznań, the small bodies of the Solar System have never been a bigger subject.

Subject of Research: Asteroids, comets and meteors in the Solar System

Article Title: The 15th Asteroids Comets Meteors Conference

Article References: Eggl, S., Kryszczyńska, A., Oszkiewicz, D., Hestroffer, D., Opitom, C., Rudawska, R., & Rożek, A. (2026). The 15th Asteroids Comets Meteors Conference. Nature Astronomy, 10(10), 1427-1428. https://doi.org/10.1038/s41550-026-02995-3

Image Credits: AI Generated

DOI: 10.1038/s41550-026-02995-3

Keywords: asteroids, comets, meteors, planetary defence, DART, Hera, sample return, meteorites, Kuiper belt, near-Earth asteroids, space surveys, ACM 2026