Select image folders or videos
Upload offline microscopy batches or online monitoring videos.
Detectron2-powered particle vision platform
MicroVision: AI-powered microparticle image analysis
Analyze images and videos with Detectron2-based instance segmentation, morphology metrics, anomaly rules, visual outputs, and downloadable reports.
Particles detected70
Detected classes2
Mean area0.00241
Abnormal count0
MicroBench images
4,300
Balanced online and offline microscopy scenes for cross-domain evaluation.
Annotated instances
148,401
Instance-level masks supporting morphology and density statistics.
Particle classes
7
Agg, block, rod, plate, bubble, droplet, and microsphere targets.
Input modes
Images / Video
Batch images and sampled monitoring videos flow into the same metric view.
Platform Workflow
From raw microscopy images to structured particle metrics
MicroVision turns microscopy image folders and sampled videos into annotated outputs, instance-level morphology tables, summary trends, and archived reports.
Configure prediction and display options
Set confidence thresholds and output visualization preferences.
Enable morphology or anomaly rules
Apply class, shape, confidence, edge, and contact checks.
Export annotated media and CSV reports
Review visual predictions, trends, metrics, and downloadable archives.
Instance segmentation
Detect and segment crystals, droplets, microspheres, bubbles, and aggregates.
Morphology metrics
Extract area, aspect ratio, circularity, diameter, and class-level summaries.
Anomaly monitoring
Apply label, shape, confidence, edge-contact, and morphology-based rules.
Run Status
No job selected
About MicroVision
Construction and Details of MicroVision
--
Website visits
--
Platform uses
Prediction Showcase
Prediction examples
Representative outputs from the deployed Detectron2 model on online and offline MicroBench-style images.
Scientific Foundation
Built from data, benchmarked for deployment
MicroVision is supported by a complete workflow from dataset construction and morphology analysis to model benchmarking, robustness enhancement, and deployable prediction.
MicroBench dataset
A unified instance-level dataset spanning online and offline imaging domains with particle-level annotations for deployment-oriented benchmarking.
Morphological space
Particle morphology is characterized by circularity, extent, eccentricity, aspect ratio, and relative area.
Model benchmarking
Seven segmentation systems were benchmarked under the same data splits and evaluation protocol.
Data scale and model selection
Benchmark results connect dataset composition, model behavior, and practical deployment choices.
Controlled Copy-Paste
Synthetic complex scenes improve robustness while preserving real-test performance.
Deployment Model
Optimal Model Selection
After systematic benchmarking, Detectron2 was selected as the deployed model because it provides strong mask accuracy, stable high-IoU performance, lower model size, and easier integration into the MicroVision web workflow. Grounded-SAM-style pipelines can produce smoother boundaries in selected cases, but their larger model size and multi-stage inference increase deployment cost.
Detailed comparisons of all models are summarized in the benchmark results section of the main text.
High segmentation accuracy
Stable high-IoU masks
Lower deployment cost
Web-ready inference workflow