Why fen.py are missed???

[francescobianco]

Fantastic work, but missing fen.py on the repository, WHY??

[kinduff]

Looks like its a non-public algorithm. you can either modify the code to get the pieces positions and generate a FEN or just use another library to convert the output chessboard into a FEN.

[francescobianco]

@kinduff how get piece position? It is require another ML layer to recognize piece? Can you show sample code?

[karunakaruna]

@maciejczyzewski, Thank you for sharing your work. Do you plan to release the piece recognition portion of the code at a later date?

I would very much like to try this for a personal project!

[Cyberdr8gon]

@karunakaruna I don't think so, I sent him an email about it and he just ignored it. It isn't like his algorithm is magical though, he is just using a modified version of this repo with a couple changes he described in the paper. If you end up implementing it, please post your results so we can verify the results in the paper.

[karunakaruna]

Thanks for that info, Sam. Will let you know if we make progress on that! -Jonny

On Thu, Dec 20, 2018 at 1:51 PM Sam Bateman notifications@github.com wrote:

@karunakaruna https://github.com/karunakaruna I don't think so, I sent him an email about it and he just ignored it. It isn't like his algorithm is magical though, he is just using a modified version of this https://github.com/jialinding/ChessVision repo with a couple changes he described in the paper. If you end up implementing it, please post your results so we can verify the results in the paper.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/maciejczyzewski/neural-chessboard/issues/2#issuecomment-449146921, or mute the thread https://github.com/notifications/unsubscribe-auth/ATrMTAx2z02qMA9DuxHQa9JbtPc-7khvks5u7AZhgaJpZM4SV0vQ .

-- Zketchbuk http://zketchbuk.tumblr.com / JonnyOstrem.com http://www.jonnyostrem.com/ / LinkedIn http://ca.linkedin.com/in/ojonnyo

[maciejczyzewski]

Our method of chessboard detection is not a modified version of Jialin Ding, because his method is MANUAL (you must provide 4 points). The main goal of our project is to provide fully automatic system (without user interaction) for chessboard detection problem (code in this repository). Solution for chess piece detection problem we treat as additional.

Nowadays, ML provides multiple solutions for classification problems (much more elegant than simple SVM), thus we don't introduce algorithm in this topic. Hoverer, we mention in our article that results of simple SVM can be boosted by introducing additional propabilities for specific chess position (using for example such projects as Stockfish), because it's obvious for example that white player cannot have 9 pawns (so maybe one is bishop?) or 3 queens. Due to possible commercial use of our 'fen.py', implementation is not presented in our article, for now we cannot publish our algorithm (but idea is described above).

@Cyberdr8gon @karunakaruna I receive a lot of junk mail asking for help in final projects/robotic arms/debugging computer vision code. It's not part of my job.

---

Citation:

---

[gilinachum]

Great project!

1. Anyone with a good reference for generating a FEN of the clipped board image?

2. I would prefer the clipped image would extend to include pieces on the edge of the board, otherwise it could be impossible to tell the exact piece when these are clipped. Example (top row pieces heads are clipped):

[linrock]

Here's open-source code that implements fen.py using ML (convolutional neural networks): https://github.com/linrock/chessboard-recognizer

It works for 2D pieces on an image of a chessboard. Could be cool to combine it with the algorithm in this paper for detecting chessboards within an image.