elucidation / chessboardfentensorflowjs Goto Github PK

Chess symbols exist in unicode, build a chessboard div and use them.

Something like

♜ ♞ ♝ ♛ ♚ ♝ ♞ ♜
♟︎ ♟︎ ♟︎ ♟︎ ♟︎ ♟︎ ♟︎ ♟︎
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
♙ ♙ ♙ ♙ ♙ ♙ ♙ ♙
♖ ♘ ♗ ♕ ♔ ♗ ♘ ♖

Right now images must be centered, well-aligned, and fill up most of the image to be detected.
Tensorflow Chessbot is able to handle non-centered boards that are smaller portions of the image.

Issues:

1. Some pieces have reversed color
2. Pawn is rendered as a bishop.

From this picture, Black pawns recognizes as Bishop : https://timkr.home.xs4all.nl/tour/fork27.GIF
And Black rook sometimes recognize as queen :(

Complete page with diagrams : https://timkr.home.xs4all.nl/tour/forks.htm

Make mobile-friendly, etc.

Since this page is served over HTTPS, now the fen-to-image url gets auto-upgraded to HTTPS (which does not work as of Feb 2024)

HTTPS: https://www.fen-to-image.com/image/32/rn1qkb1r/p4ppb/1pp1pn1p/4N3/2BP2P1/1QN1P2P/PP3P2/R1B2RK1.png
HTTP: http://www.fen-to-image.com/image/32/rn1qkb1r/p4ppb/1pp1pn1p/4N3/2BP2P1/1QN1P2P/PP3P2/R1B2RK1.png

Mixed Content: The page at 'https://www.tetralark.com/ChessboardFenTensorflowJs/' was loaded over HTTPS, but requested an insecure element 'http://www.fen-to-image.com/image/32/rn1qkb1r/p4ppb/1pp1pn1p/4N3/2BP2P1/1QN1P2P/PP3P2/R1B2RK1.png'. This request was automatically upgraded to HTTPS, For more information see https://blog.chromium.org/2019/10/no-more-mixed-messages-about-https.html

Replace this with something else.

I am trying to find the FEN of https://i.redd.it/jafm8w43mzk31.jpg
I used the Live Demo of FENbot, and it gave me the correct FEN. However, when I do:

$ ./tensorflow_chessbot.py --url https://i.redd.it/jafm8w43mzk31.jpg

I get the incorrect FEN. Why does this difference happen?

TF model takes time to load but user can upload an image via button before that.

Am working on a Chess teacher

Chesspieces are from https://commons.wikimedia.org/wiki/Category:SVG_chess_pieces/Maurizio_Monge

With extra dropshadows showing defense/attack

This is the 1997 Kasparov vs Deep Blue rematch game #1 (Deep Blue resigned after G7, but won the 6 game match)

Beta version at: https://chessly.github.io/?fen=4r3/8/2p2PPk/1p6/pP2p1R1/P1B5/2P2K2/3r4

Input is:

Wrong output is:

• Black King becomes White King
• White King becomes White Pawn
• White Bishop becomes White King (second one on the board)
• White Rook undetected
• Black Pawn with 2 dropshadows becomes Black Queen

HTH

Implement the ChessboardDetect algorithm to find boards from images.

especially, the white queen, and some other obvious pieces

white queen detected as black queen.
pawn as bishop and so on.

Hello! This is such an excellent project idea!

In case it's helpful, here's an image that wasn't parsed correctly:

What it came up with: R1PKQP1R/2P1PP1P/P1NP1N2/1P1p2K1/b3p3/2n2n2/ppp2ppp/1kr1q2r

As you will see, it messed up the bishops on the 8th rank, and misidentified one of the black bishops as an extra white king