The stakes are higher than they might appear. Grain morphometry underpins a remarkable breadth of Earth science. The sizes and shapes of clasts record how sediments are produced on hillslopes, how rivers hydraulically sort and transport their loads, and how deposits in the stratigraphic archive were laid down. Grain size data have even been used to reconstruct transport conditions on Mars, where the shapes of ancient pebbles hint at how far water once flowed. But the traditional ways of collecting these data are laborious and biased. Manual Wolman counts and line sampling yield relatively few observations and depend heavily on the observer. Point-cloud methods struggle with grain occlusion and require geometric assumptions. Earlier image-based tools either need field calibration or systematic manual correction, and they tend to over- or underestimate grain sizes in predictable but troublesome ways.

Machine learning promised to change this, and to some extent it has. Texture-based neural networks can predict summary statistics of a grain size distribution directly from image texture, while segmentation-based approaches delineate every grain individually, opening the door to datasets of tens of thousands of measurements. The catch, as Mair and colleagues emphasize, is that these segmentation models have behaved like narrow specialists. Trained on homogeneous, task-specific datasets, they lose accuracy the moment they encounter a new camera, a new lighting regime or a new sediment type. Each new setting has effectively demanded a new model and a freshly curated training set, which has blunted much of the promised efficiency.

The Bern team’s solution was to reach for a different class of model altogether. In recent years, transformer-based foundation models such as Meta’s Segment Anything Model, or SAM, trained on eleven million images and more than a billion annotation masks, have shown an extraordinary ability to generalize to unseen data. But SAM in its basic form is inefficient at densely segmenting large numbers of similar objects, which is precisely what a gravel bar photograph demands. The breakthrough came with Cellpose-SAM, a next-generation model developed for segmenting cells in biomedical imagery. It grafts SAM’s transformer image encoder onto the flow-based segmentation machinery of the original Cellpose framework, which tracks predicted vector flows to separate touching objects. The hybrid keeps SAM’s generalization while excelling at dense, prompt-free segmentation of many instances of one object class, and its architecture has grown to more than 304 million trainable parameters, roughly fifty times larger than the U-Net backbone it replaces.

Retraining such a model for geoscience required data, and lots of carefully annotated ones. The researchers assembled the ImageGrains 2.0 dataset, combining 81 previously labelled tiles with 162 new ones for a total of 243 image tiles containing more than 29,000 manually annotated grain masks. The imagery spans fluvial gravels photographed from uncrewed aerial vehicles and handheld cameras in Switzerland, Spain, Peru, New Zealand and Namibia; painted and unpainted pebbles used to track grain mobility; angular proglacial deposits; near-vertical outcrops of lithified conglomerates; X-ray computed tomography slices of glacial till; and micro-CT images of bioclastic marine sand. Crucially, the team deliberately included awkward content, such as vegetation, sieves, hands, scale objects, water bodies and harsh shadows, to test whether the model could distinguish genuine grains from everything else in a messy field photograph.

The performance gains are substantial. Benchmarked against the manual annotations using average precision at an intersection-over-union threshold of 0.5, the fine-tuned Cellpose-SAM model achieved a median score of 0.72 across all image tiles, roughly 20 percent higher than the second-best method, a Cellpose 2 model trained on the same data. The advantage held in both training and unseen test splits, and it proved largely independent of how many images each sediment type contributed to training. The model scored highest of all on X-ray CT tiles, which made up only about a tenth of the dataset. Even more telling was the generalization test: when the researchers withheld two entire subsets from training, the model still matched or beat specialist models that had been fine-tuned on those very images, evidence that the transfer-learning strategy preserves the broad visual competence of its SAM heritage.

Accurate masks translate directly into accurate measurements, and here the numbers are striking. The model recovered 93 percent of ground-truth grains, and mean differences in grain size were around 3 percent or less for both the long and short axes of ellipse fits to each grain. The resulting grain size distributions were statistically indistinguishable from the ground truth in 88 percent of cases for the a-axis and 82 percent for the b-axis, compared with only 54 and 57 percent for the second-best model. Shape metrics fared equally well, with average deviations below 2 percent for roundness, roughness, elongation and orientation. The team also identified a practical rule of thumb: perfect statistical agreement between predicted and true grain size distributions was only achieved when segmentation scores exceeded roughly 0.68 average precision, a threshold that varied with how well sorted the sediment was.

The framework extends into three dimensions as well. Applying the model to a stack of 400 X-ray CT slices from a drill core of glacio-fluvial sediment in southern Germany, the researchers segmented 4,647 well-defined coarse grains in full volumetric detail, something 2D photographs can never provide. Because CT scans capture complete grains rather than partially occluded surfaces, they sidestep the geometric extrapolations that plague LiDAR and photogrammetry-based approaches. The authors note that rigorous 3D benchmarking still awaits a manually annotated volumetric ground truth, whose creation has so far been impractical, but the demonstration suggests CT-based sediment analysis, from clast fabric to depositional history, is now within reach of automated workflows.

Accessibility was a design priority. ImageGrains 2.0 ships as an installable Python package with an interactive graphical user interface for annotating images and correcting model predictions, so researchers need not write code at all. Training runs in under ninety minutes on a high-end GPU, and although transformer models are demanding, the authors show that fine-tuning remains feasible on a mid-range desktop graphics card, with inference requiring as little as 3 gigabytes of memory. Because the model builds on already-trained foundation weights, its marginal energy footprint is modest compared with training such architectures from scratch. Adapting the model to a genuinely new sediment type or imaging style can require as few as seven additional annotated tiles, a dramatic lowering of the barrier that once made every new site a modelling project of its own.

The deeper message of the study concerns where progress will come from next. The authors argue that future gains in grain measurement will depend less on new architectures than on larger, more variable, publicly shared annotated datasets, ideally with annotations from multiple experts to establish consensus-level benchmarks. They explicitly frame their work as a step toward a foundation model for granular particles in geoscientific imagery, one that could serve any discipline needing dense segmentation and morphometric analysis, from sedimentology to planetary science. With the dataset, model weights and code all openly available, the Bern team has effectively invited the community to build on their groundwork, turning what was once a tedious field chore into a problem that a well-fed neural network can now handle in minutes.

Subject of Research: Deep-learning-based automated segmentation and morphometric measurement of sediment grains in geoscientific imagery

Article Title: ImageGrains 2.0: Improved precision and generalization for grain segmentation

Article References: Mair, D., Witz, G., Do Prado, A., Garefalakis, P., Wild, A., Ville, F., Schuster, B., Horn, M., Österle, J., Fabbri, S. C., Litty, C., Achleitner, S., Leistner, S., Hiller, C., & Schlunegger, F. (2026). ImageGrains 2.0: Improved precision and generalization for grain segmentation. Earth Surface Dynamics, 14(4), 527-551. https://doi.org/10.5194/esurf-14-527-2026

Image Credits: AI Generated

DOI: 10.5194/esurf-14-527-2026

Keywords: ImageGrains 2.0, grain segmentation, Cellpose-SAM, deep learning, sedimentology, grain size distribution, X-ray computed tomography, transfer learning, Segment Anything Model, fluvial sediment, geomorphology, open-source software

A new study published in BMC Medical Imaging by Suxiang Yu, Xiaoqin Yang, Yihe Duan, Dun Hua, Bai Yun, Huimiao Sun, Lingling Zhang, Feihong Wu, Dandan Yang, and Xin Huang takes aim at exactly this problem. The team, spanning pathology departments and engineering schools in China, evaluated and adapted Cellpose-SAM, a modern generalist cell segmentation model, to the specific and stubborn challenge of instance segmentation in overlapping cervical cytology images. Their results, while modest in absolute terms, point to a practical lesson for the field: even powerful foundation models need careful, task-specific tuning before they can be trusted on the messy realities of clinical slides.

Cellpose-SAM itself represents an interesting convergence of two research threads. Cellpose, originally released in 2020, popularized a two-step approach to cell segmentation in which a neural network first predicts the flow of gradients from every pixel toward the center of its parent cell, and an algorithm then follows those flows to group pixels into individual instances. This design elegantly sidesteps the need for the network to label each cell with a unique identifier, which is notoriously difficult to train. The SAM in the name refers to the Segment Anything Model, a vision foundation model trained by Meta AI on an enormous corpus of natural images. By marrying SAM’s pretrained visual backbone to Cellpose’s flow-based decoding, Cellpose-SAM inherits strong generalization to cell types and imaging modalities it has never seen before, including microscopy styles far removed from its training distribution.

But strong generalization is not the same as optimal performance on any given task. As the authors note in the study, the default configuration of Cellpose-SAM may not be suited to densely overlapping cervical cytology images, where cell adhesion, weak cytoplasmic contrast, and ambiguous boundaries conspire to obscure individual cell structures. When two overlapping epithelial cells stain nearly identically and share a barely perceptible junction, the model must decide where one cell ends and the other begins, a judgment that even the underlying flow field may represent poorly without further training on comparable data.

To give the model that training, the researchers assembled a single-center retrospective dataset from the Fourth Central Hospital of Baoding in Hebei Province, China. The dataset consisted of 433 images of overlapping cervical cells, each accompanied by expert manual annotations that served as the ground truth against which the algorithm’s outputs were measured. These images were divided into training, validation, and independent test sets at a ratio of 70 percent, 10 percent, and 20 percent, respectively. The split is a standard but important design choice: the training set teaches the model, the validation set guides tuning decisions, and the test set, held completely apart from both, provides an honest estimate of how the adapted system performs on data it has never encountered during development.

The adaptation strategy had two complementary components. The first was fine-tuning, in which the pretrained Cellpose-SAM weights were updated on the cervical cytology training images, allowing the model’s internal representations to adjust to the specific staining patterns, cell morphologies, and clutter of Pap-smear preparations. The second was inference-parameter optimization, a step that is often overlooked but can be as consequential as retraining itself. Cellpose-SAM exposes several knobs at prediction time: the expected cell diameter, which calibrates the scale at which the network looks for objects; the flow threshold, which determines how strictly the gradient-following algorithm trusts its flow predictions when assembling cell instances; and the cell probability threshold, which decides how readily the model declares a pixel to belong to a cell at all. Rather than accepting defaults, the team systematically tuned these parameters on the validation set, using it as a proxy for the true target distribution and then freezing the chosen values for evaluation.

Performance was assessed on the independent test set using four established segmentation metrics. The Dice coefficient, which measures the harmonic mean of overlap between predicted and annotated regions, reached 0.8737 for the adapted model. Intersection over Union, a stricter spatial-overlap measure, came in at 0.7199. The Aggregated Jaccard Index, a metric designed specifically for instance-level segmentation because it penalizes both missed and spurious instances across a whole image, registered 0.7024. The error rate, which captures instances the model failed to segment correctly, was 0.1551. The authors performed statistical comparisons between the default and adapted configurations using paired image-level analysis, and the adapted model showed improved performance across these measures, suggesting better agreement with the expert manual annotations in the overlapping-cell regions.

The numbers deserve careful interpretation. A Dice coefficient of roughly 0.87 indicates substantial pixel-level agreement, and an IoU above 0.72 is respectable for a task where the ground truth itself is contested by virtue of ambiguous cell boundaries. The AJI of about 0.70, meanwhile, is the more sobering figure, because it aggregates performance at the level of whole instances: every cell the model misses entirely or invents from nothing drags the score down across the entire image. An error rate of about 15.5 percent means that roughly one in six or seven predicted instances does not match the annotation well. In other words, the adapted Cellpose-SAM is a capable assistant rather than a finished autonomous cytologist, and the gap between 0.87 pixel overlap and 0.70 instance accuracy quantifies precisely the residual difficulty of separating cells that touch and overlap.

What makes the study noteworthy methodologically is its insistence that off-the-shelf foundation models are starting points, not endpoints. The cervical cytology community has seen a wave of deep-learning screening tools, many of which treat segmentation as a solved problem and focus instead on classification of abnormalities. Yet if the segmentation stage silently merges overlapping cells or slices one cell into two, every downstream measurement, from nucleus-to-cytoplasm ratio to nuclear area, inherits that distortion. The Baoding team’s finding that simply tuning the cell diameter, flow threshold, and cell probability threshold on a small validation set yielded measurable gains over defaults suggests that laboratories adopting Cellpose-SAM or similar tools should budget time and annotation effort for exactly this kind of calibration, tailored to their staining protocols and scanner optics.

The authors are appropriately measured about the limits of their work. The dataset comprised 433 images from a single center, which raises questions about how well the adapted model would transfer to slides prepared with different stains, scanned on different instruments, or drawn from patient populations with different distributions of inflammation, atrophy, and glandular cells. They state explicitly that further external validation and broader benchmarking are required before clinical generalization can be established. The retrospective study was approved by the hospital’s ethics committee under approval number 2022011, conducted in accordance with the Declaration of Helsinki, and used de-identified, anonymized images, with the informed-consent requirement waived for the retrospective design. The work was funded by the S&T Program of Hebei under project number 22377774D, and the authors declared no competing interests.

Still, the trajectory is easy to read. Foundation models pretrained on millions of natural images are rapidly becoming the default starting point for biomedical image analysis, and the interesting research questions are shifting from how to build such models to how to adapt them efficiently to narrow, high-stakes domains. Cervical screening is a particularly attractive target: it is a globally distributed task with a persistent shortage of trained cytologists, and the cost of a missed abnormality is measured in human lives. If modest fine-tuning and parameter tuning can push a generalist segmentation model to useful accuracy on the notoriously tangled cells of a Pap smear, the path from laboratory benchmark to screening-room assistance becomes considerably more concrete, and the 433 clumped, ambiguous, stubbornly overlapping cells of Baoding become a small but meaningful waypoint on that road.

Subject of Research: Instance segmentation of overlapping cervical cells in Pap-smear images using adapted Cellpose-SAM

Article Title: Adapting Cellpose-SAM for instance segmentation of overlapping cervical cells in Pap-smear images

Article References: Yu, S., Yang, X., Duan, Y., Hua, D., Yun, B., Sun, H., Zhang, L., Wu, F., Yang, D., & Huang, X. (2026). Adapting Cellpose-SAM for instance segmentation of overlapping cervical cells in Pap-smear images. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02845-8

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02845-8

Keywords: Cellpose-SAM, cervical cytology, Pap smear, instance segmentation, deep learning, fine-tuning, medical imaging, computer vision, cervical cancer screening, parameter adaptation, BMC Medical Imaging, Adapting