A small Python app that guesses cursorwritten digits from 0 to 9. -> The app uses a simple manually trained dataset , NOT mnist. You draw a digit, tell the app what digit it is, and it saves that example. After enough examples are trained, the app can guess new drawings.
Fixed what the review lacked , now the project comes in exe file !
update : 12/06/2026 ( few hrs after the previous devlog )
i updated the sample size to 200+ , my goal is to make the default dataset json file in github near about 1000 , 100/digit , so its even better , for now i am satisfied and will see tomorrow for improvements/data training!
update : 12/06/2026
i was working with the mnist dataset and the tensorflow method to create this , and my model trained well ( 0.91 accurancy 0->1 ) , but it never worked during testing , so i gave up the mnist set
INSTEAD… i tried to make something new with the help of codex and power of json !
introducing the NEW and IMPROVED visual digits !
( github+FULL source code for PROPER info ( you won’t regret if ur interested )
in a nutshell tho :
now you can train it yourself , it currently has 10 samples / digit but you can now train it on your own system and even show it to me through github if your training dataset is superior !
it might be weaker than the CNN pathway but its simpler in working so yeah !
code is complete , i took help from readymade projects in this same field , im not an expert in AI since i never learnt it in school , just basic python so i needed alot of help from resources regarding this project.
10 epoches left a lot of room of error , im currently training my model with 20 epoch so its more accurate , will see if it needs more training soon.
( excuse if step times are slow my laptop is very mediocre , had to close everything for optimal ram usage )
