Ballooned drawing
PDFEvery dimension is numbered and connected to its feature. Export a ballooned PDF or image and print the sheet the inspector works from.
Features
One chain from drawing to delivery
Import, detect dimensions, review OCR, balloon every characteristic and export the deliverables — all inside one Windows workstation. The sections below follow the order of the actual workflow.
- PDF / PNG / JPG import
- 1280+ detection models
- Windows 10 / 11
- Detection runs locally
Workstation: drawing list on the left, canvas in the middle, results and exports on the right
Deliverables
What you hand over: three files
Recognition only matters when it becomes a file you can inspect, archive and send to a customer. All three are generated from one review pass, with configurable fields and templates.
Inspection detail
ExcelIndex, inspection item, feature description, nominal, tolerances and limits written into a mapped Excel sheet, ready to edit or import.
Inspection report
ReportWrite results into your customer Excel template. A built-in template is included and header, detail area and field columns can be detected automatically.
Results are managed per page; numbering matches the canvas
Capabilities
Four groups, in workflow order
Each group maps to one concrete step of delivery. Modules you do not need stay out of the way; the ones you do need are all here.
Detect & recognise
A YOLO vision model finds every dimension annotation first, then reads the text.
Dimension detection
A YOLO vision model frames every dimension annotation on the whole drawing — no manual region selection — with confidence presets and 0.05 fine tuning.
Missed-item scan
Scan for likely missed regions after detection — edges, dense annotation blocks and vertical text.
AI text recognition
Read dimensions, tolerances, surface finish and GD&T symbols, switching OCR models per drawing type. Cloud OCR uses the DeepSeek vision model.
Tolerance standards
Built-in GB/T 1804-m and GB/T 1800.01 IT10 standards fill general tolerances after recognition.
- Slice experiment
- CPU / GPU CUDA / DirectML
- Manual boxes
- About 1s per drawing
- Cloud OCR: DeepSeek vision model
Review & correct
Everything is confirmed by a person before it becomes a deliverable.
OCR re-check
Re-process failed or suspicious items, or correct a value directly in the table.
Result table
Index, item, feature description, nominal and tolerances are editable, with warning rows highlighted.
Display switches
Boxes, balloons, numbers, leader lines and the grid can be toggled without touching the data.
Manual boxes
Frame a missed annotation by hand and recognise it; tighter boxes give more stable results.
- Canvas ↔ table sync
- Rotate 90° and retry
- Symbol palette
Balloon & export
Turn confirmed results into files you can hand over.
Balloon numbering
Balloons are auto-grouped and numbered per view with sequential numbering; balloons and leaders avoid dimension lines so annotations stay unobstructed, and positions can still be nudged by hand.
Ballooned output
Export ballooned PDF or images with page selection, batch printing and overlay switches.
Excel detail
Export the structured detail sheet with configurable columns, ready for statistics or existing systems.
Inspection report
Write results into a customer Excel template, with AI-assisted template and column mapping setup.
- Batch balloon printing
- Packaged export
- Per-drawing selection
Batch, revisions & models
Whole sets, frequent revisions and specialised drawings each have a path.
Batch processing
Detect, review, number and export across files and pages in one run.
Revision comparison
Old and new drawings side by side; added, removed and changed dimensions are listed.
Model training
Save corrected results as samples, generate a training script and train a dedicated model.
Model market
Search, download and enable shared ONNX models, or upload your own for the team.
- History and project files
- Custom model import
- 1280+ models included
Recognition
Locate first, read second
Plain OCR can read the text but does not know where the number sits on the drawing. TuMo Vision detects the annotation box first, then binds each recognised value to its location — that is what makes numbering and detail sheets meaningful.
- Detection boxes carry drawing coordinates
- Each value is bound to one box
- Canvas and table locate each other during review
Recognition logic: locate the box, then read the text
Batch
Run the whole set, not one drawing at a time
Batch jobs run across files and pages with unified detection, review, numbering and export. Batch balloon printing lets you choose pages, scale and overlays so delivery keeps moving.
- Mixed multi-page PDF and images
- Batch detect / number / print
- Resumable jobs, results saved per page
Batch balloon printing: pages, scale and overlays
Revisions
Stop hunting for changes by eye
Load the old and new drawing and run the comparison. Added, removed and changed dimensions are listed with status, match confidence and tolerance differences; each row locates the annotation on the drawing.
- Synchronised side-by-side zoom
- Change list with status and confidence
- Export the change list for review
Old and new side by side, every change listed
Model market
Start from a model, not from zero
1280+ industry detection models ship with the client and work right after install. For a more specialised drawing style, search and enable a shared model in the market, or import your own trained model.
- Browse, search and enable models in the client
- Upload your own ONNX models
- Switch models per drawing type without changing the workflow
Model market: search, enable, or upload your own
Boundaries
The prerequisites, stated up front
Learning a boundary during a demo costs more time than reading it here. Four things worth knowing before you start.
Read the data & security pageLocal first
Detection, ballooning and export run in the Windows client; drawings do not have to be uploaded first.
Cloud OCR
Cloud OCR uses the DeepSeek vision model, so it needs a network connection and a signed-in account, metered by plan.
Human review
Results are confirmed by a person, especially critical dimensions. Software assists inspection, it does not replace the decision.
2D drawings
Works on 2D engineering drawings in PDF or image form. It is not a 3D reconstruction or a CAD editor.
Run one of your own drawings through it
The installer is self-contained — no .NET install needed. Within ten minutes of downloading you can see your first ballooned drawing and inspection report.