How do you detect the pieces? about chessvisionbot HOT 6 CLOSED

CPU is fine, just slower

from chessvisionbot.

Hey Peter, the pieces for the screenshot are in the folder 'pieces'. But I don't use these for detection but only for visualization of the detected position. For the detection I use a trained CNN so it works with basically every set of chess pieces as long as they are at least similar to the more standard ones. The detection is done in bord_basics.py At the moment there is a lot of code in these because I'm playing around with caching the detections, so I don't have to redetect the same piece that hasn't moved since the last detection.
There are two variations of detection. One is the Binary-Detection model, which just evaluates if a square is empty or not. I use that model to update the position when I already detected the position completely. Then there is the Class-Detection model, this model can distinguish between all the pieces and colors to build the initial position. You can find the models in their respective folders. They are first initialized when you start the application in ml_model.py

from chessvisionbot.

That is really interesting. I tried the pyautogui.locateOnScreen but i fould it unreliable. A CNN is probably a solution. Is your classifying CNN able to run on CPU or do you need a proper graphics card?

from chessvisionbot.

Also, I am just curious, is there a source code of this tensorflow model in this repository? I am working on a similar program to yours and I want to be able to really look under the hood.

Regards.

Peter

from chessvisionbot.

As I said the models are in their respective folders. I even put an image of their structure into the folder. If you want to examine them further feel free to load them into keras and analyze them

from chessvisionbot.

Oh. I am really sorry. Thanks for the help!

from chessvisionbot.