Gaz.AI | Latest Screenshots | Quick start (Windows, Mac, Linux) | What it does | Privacy statement | News

Gazelle 22 - a new image recognizer

Team built? [More news]

Latest Screenshots

Gazelle is learning to read.
(All characters shown are drawn by Gazelle trying to mimic a screenshot.)
Not all that bad already!
And it wasn't even trained on Courier.
And yes, that fancy artwork could be considered a bug.


Which shape is similar to a 5?


Gazelle's cheat sheet for all shapes in the extended latin alphabet


Working on a compressed image format (and the demo is a beautiful main program too)


Recognizing Toilet Paper [Video]


A more evolved data graph


First data graph


Finding topological graph mappings


A nice, straight-forward motion detector


Segmenting an image by covering it with single color tiles
(left: input, right: output scaled up)


Semi-automatic conversion of left-arrow scripts to Java(X)


Optimized script compilation time to below 1 ms (displayed value is rounded up!)


PDF Import (commercial feature)


Learning fonts!


Training the computer as a math student... :)


Finding junctions and line ends in the letter skeletons


Artifact-free reduction of letters to 1 pixel-wide skeletons!


Finding a polygon's corners (marked red)


A 5 point gradient is recognized!


Rendering some multi-point gradients [gradient detection coming up too]


Some recognition of... something!? LOL


Checking in how many fonts (out of 88) a letter is actually different.
Turns out "e" has the most versions (83) and "l" (lowercase L) has the fewest (76).


Theoretical number of different images at 3*3 pixels and 3 colors


Each letter reduced to 3x3 pixels and 3 colors. Can you still tell them apart? (Each letter in the grid is an H, e, l or o.)


Each letter was padded into a square


Extracted letter matrix ready for learning


Segmenter usually finds 5 letters as expected (the word is 'Hello'), but a few fonts still confuse it.


Project story editor with hyperlinks & auto-run mechanism for user scripts


Gazelle has split a word into invididual letters.


This screenshot shows my cute new scripting language which is very flexible.


Here you see Gazelle's preprocessing filters applied to a screen cam.


Quick start (Windows, Mac, Linux)

1. Install Temurin 17 or any recent Java version. (Their smaller "JRE" version is fine too. If you already have Java, just go to step 2.)

2. Save and double-click Gazelle.jar (it's under 30 MB).

What it does

Gazelle 22 watches your screen and recognizes things.
It is being developed right now by Stefan Reich at Gaz.AI. Contact.

Source repository. Older releases on GitHub.

Privacy statement

Gazelle 22 runs locally on your machine.
It does not upload any data to any server.
Server connections are made only for downloading program parts and checking for updates.
All code is open source.


Team built? [2022/9/3]

Post on AGI list [2022/8/21]

Looking to build a new team to finish the Shape Collector [2022/8/16]

Team grows! Welcome Avinash & Viwekaditya :) [2022/5/5]

A new team is formed!! Welcome Rohit and Pratik!! [2022/4/23]

Alright... let's get into a fresh productivity rush! [2022/4/8]

I will now convert a bunch of fonts to topological meshes and create the first truly flexible text recognition algorithm. [2022/3/28]

I have a cute idea of how to recognize letters using scaled&posterized images (see this screenshot) and will try to implement it now. [2022/3/10]

Haven't been posting updates, but I am actively working on Gazelle. A first text recognizer should appear in a few days. [2022/3/3]

Loads of new stuff. You can paint in Gazelle now, just like with MS Paint. A mini video player is also integrated, and updates are faster now. [2022/2/6]

Gazelle handles multiple databases now [2022/2/1]

Post on AGI list [2022/1/29]

Another release!! This one can actually do a first simple recognition. If you type in the code from the screenshot... ^^ [2022/1/29]

Put new release on home page [2022/1/28]

News post [2022/1/22]

News post [2022/1/7]

Image recognizers should return mathematical proofs [2021/9/10]

Gazelle looks at its own logo and paints it red [2021/8/20]

Imprint: Stefan Reich, Dubenhorst 27, 25474 Ellerbek