SCIENTIFIC IMAGE AI FOR CT & SEM
United Vision
Segment, enhance and identify objects in scientific CT images (including micro-CT and XRM) and SEM images (including FIB-SEM serial sections). Use open-source models without training, or fine-tune and retrain supported models on your own data.
01 / APPLICATIONS
Bring segmentation, enhancement and object recognition into one workflow for structures and objects in scientific images.
CT / micro-CT / XRM
Work with CT, micro-CT and XRM images and slice stacks to segment and identify structures such as pores, particles and material phases. Combine image enhancement with 3D viewing to inspect the results.
IMAGE SLICES · VOLUMES · INTERNAL STRUCTURES
SEM / FIB-SEM
Work with SEM surface and cross-sectional images, plus serial-section data such as FIB-SEM, for microstructure segmentation, image enhancement and object recognition. Inspect results in 2D images and image stacks.
SURFACES & CROSS SECTIONS · SERIAL SECTIONS02 / CAPABILITIES
Choose methods for segmentation, enhancement and object recognition, then take the results into measurement and structural analysis.
Separate objects from background and extract regions or individual instances. Combine automatic and interactive segmentation to inspect structural boundaries in scientific images.
SEMANTIC · INSTANCE · INTERACTIVE SEGMENTATIONUse denoising and super-resolution tools to improve image appearance. Compare input and output in the preview to support further observation and analysis.
DENOISING · SUPER-RESOLUTIONLocate objects with detection models or distinguish image categories with classification models. Apply trained models to new images in batches.
OBJECT DETECTION · IMAGE CLASSIFICATION · BATCH INFERENCELoad images, serial sections and volumes. Inspect structures in the integrated 3D viewer, with inputs including TIFF, HDF5 and VGL, or connect to Dragonfly for further analysis.
IMAGE STACKS · VOLUME VIEWING · DRAGONFLY INTEGRATIONVISUALIZE THE PROCESS
Explore the relationship between image noise, visible detail and region labels.
Interactive illustration · Synthetic shapes, not real CT / SEM results
03 / OPEN-SOURCE MODELS
Use open-source models and choose an approach based on your images, annotations and research task.
01 / NO TRAINING REQUIRED
Choose a pretrained model that supports use without training. Run inference directly or guide interactive segmentation with points and boxes. Check results on your own CT / SEM images before deciding whether training is needed.
02 / FINE-TUNE & RETRAIN
Prepare images and annotations to fine-tune or retrain supported models for your specimens, imaging conditions and target classes. Validate on separate data before applying the model to batch analysis.
The model library includes open-source models such as SAM, Cellpose and StarDist. Image suitability, training-free use, fine-tuning support and license terms vary by model.
04 / WORKFLOW
Load CT / SEM images or serial sections. Add annotations when training is needed.
Use a model without training, or fine-tune and retrain on your own data.
Run segmentation, enhancement or object recognition. Inspect image previews and 3D views.
Save outputs, run automated analysis or connect to Dragonfly for further exploration.
05 / ANALYZE & CONNECT
An integrated particle workflow brings segmentation, measurements and reporting into repeatable steps, with statistical tables and Word reports.
Connect to Dragonfly 3D World to transfer images and annotations, and run reusable analysis recipes in your connected session.
Dragonfly integration requires a separate installation, valid license and connection. Model capabilities depend on the selected model, dependencies and hardware.
06 / GOOD TO KNOW
Scientific images, with a focus on CT (including micro-CT and XRM) and SEM (including FIB-SEM and other serial sections). Choose segmentation, enhancement, detection and classification methods to suit your data.
No. Models that support training-free use can run directly or with interactive prompts. If an existing model does not suit your images, prepare annotations to fine-tune or retrain a model that supports training.
This is an introduction to the desktop application. Analysis happens in the desktop software. This website does not upload or process your research images.
Designed for 64-bit Windows 10 / 11. A CUDA-capable NVIDIA GPU is recommended; some features can use the CPU. Requirements vary by model and dataset size.
This first version introduces the product. Public downloads and online purchases are not available here yet. Distribution and licensing information will be added later.
Model training, inference and local particle analysis are available in the desktop workspace. Dragonfly integration is an additional workflow requiring its own installation and license.
United Vision / VyZ.World
Connect open-source AI models with real analysis tasks in CT and SEM research imaging.