About

Finishing the delivery chain

TuMo Vision is a Windows desktop application built by Jiuzhang Arithmetic Studio: import an engineering drawing, then export a ballooned PDF, an Excel detail sheet and an inspection report in one chain — without moving data between tools.

At a glance
Product
TuMo Vision
Built & maintained by
Jiuzhang Arithmetic Studio
Runs on
Windows 10 / 11
Stage
Early product, iterating continuously
Why

Recognition is not the slow part — moving the data is

Manufacturing and quality teams are not short of drawings: PDFs on the file server, scans in a chat window, revision notices in email. What costs time is turning them into something you can hand over — copying dimensions into Excel, ballooning by view, and laying out a ballooned drawing and report that someone can sign.

Generic vision models can frame a few annotations, but they do not know that an item belongs in an inspection table or that its tolerance follows the GB/T 1804 default grade. So TuMo Vision was aimed at the deliverable from the start: not a pile of coordinates, but three files an inspector can work from and a customer can sign.

That is also why detection runs locally, cloud features are explicit options and prices are written on the page. Those choices make the product slower to build, and a better fit for the real workplace.

The value of the software is not the moment of recognition, but how much data movement it removes afterwards.

Our position
TuMo Vision workspace with a loaded engineering drawing

A Windows workstation, ready after install

TuMo Vision inspection workstation
Capabilities

What the software actually does

Five steps, one chain. Every step keeps a reviewable state instead of producing a one-shot black box.

  1. 01

    Import

    PDF / PNG / JPG, multi-page and batch import with page management.

  2. 02

    Detect dimensions

    A YOLO vision model frames every annotation on the whole sheet, running on local CPU / GPU inference.

  3. 03

    Review OCR

    Extract values, correct them item by item and filter by confidence; cloud OCR uses the DeepSeek vision model.

  4. 04

    Number balloons

    Balloons are grouped per view, numbered sequentially and automatically avoid dimension lines, while staying tied to features.

  5. 05

    Export

    Ballooned PDF, Excel detail sheet (GB/T 1804 default tolerance) and inspection report in one run.

Beyond the core

Batch processing

Detect, review, number and export across files and pages in one job.

Revision compare

Compare dimensions and tolerances between revisions, export the change list.

Training loop

Corrected results become samples that fit your drawing types.

Model market

1280+ industry detection models, swappable per drawing type.

Principles

Three principles we will not trade away

Features can be added over time. These three decide what the product looks like.

01

Deliverables first

A feature earns its place by making the delivery files earlier and more accurate. A perfect recognition result that still needs manual rework has not solved anything.

  • Ballooned PDF, Excel and report are first-class output
  • Templates and field mapping are configurable
  • No coordinate-only demos
02

A clear data boundary

Where a drawing is processed, whether it is uploaded and whether it is metered should be visible and your decision.

  • Local detection and export
  • Cloud OCR metered per run
    03

    Pricing said plainly

    The free quota, monthly and annual prices are all on the site — no "contact us to hear the price" tier. Team licensing and private deployment are quoted on real scale, with the scope spelled out first.

    • 200 free detections
    • ¥49 / month, ¥490 / year, in the open
    • No hidden usage clauses
    Boundaries

    What we do not do

    Writing down what we cannot or will not do is more useful than another claim of being industry-leading.

    We do not promise 100% automation

    Drawing quality, annotation density and scan clarity vary widely. The review step stays in the flow to absorb that uncertainty.

    We do not upload drawings silently

    Detection and export run locally; cloud OCR requires you to choose it for a job; uploading for training is a setting.

    We do not fake customers or certificates

    No unauthorised logos, testimonials or certificates appear on this site. Ask for references and we will share real trial records.

    No online checkout, for now

    There is no online payment on the site. Memberships are activated by a person confirming the account and usage scale, so you do not pay for something that cannot run your drawings.

    How to start

    Start with one drawing, in four steps

    You do not need to complete procurement to begin: validate with the free quota, then decide on a membership or team licence.

    Install

    A Windows 10 / 11 installer. Import a drawing right after installation; no runtime setup.

    Validate on your drawings

    Run the full chain on the drawing type you handle most: detect, review, number, export.

    Choose licensing

    Monthly or annual for individuals; team licensing quoted by device and concurrency; cloud OCR metered separately.

    Keep iterating

    Models and features keep updating, and corrected results can become samples for your drawing types.

    The 200 free detections are enough for one full delivery trial. If you need more time to evaluate, tell us.

    Contact

    The people who build it also maintain it

    Questions reach the people who make the product. Drawing types, delivery formats and deployment environments are all fair game.

    Ten minutes to see one drawing through

    No environment setup. Take a drawing you know well, go from import to exported balloons, and judge it yourself.