- Blind mode tutorial
lichess.org
Donate

Digitize OTB Games with ChessCam

#22

@computer_trash said in #20:

@BlindfoldBlunderer this is super cool! do you have any information you could provide on the accuracy of the software? Not trying to rag, just a stats nerd and am aware no CV model is perfect

Confusion matrix of object detector performance - https://imgur.com/a/Oudf6oE

Bar chart of tracker performance - https://imgur.com/a/ErgOCgU
(Each dataset is a set of source videos, a score of X means that X% of the video was tracked correctly before a mistake was made)

The results + people's experiences have led me to believe that if you can arrive on a "good" setup for your camera, you can get 95%+ accuracy no problem :)

@computer_trash said in #20: > @BlindfoldBlunderer this is super cool! do you have any information you could provide on the accuracy of the software? Not trying to rag, just a stats nerd and am aware no CV model is perfect Confusion matrix of object detector performance - https://imgur.com/a/Oudf6oE Bar chart of tracker performance - https://imgur.com/a/ErgOCgU (Each dataset is a set of source videos, a score of X means that X% of the video was tracked correctly before a mistake was made) The results + people's experiences have led me to believe that if you can arrive on a "good" setup for your camera, you can get 95%+ accuracy no problem :)
#23

@computer_trash said in #20:

@BlindfoldBlunderer this is super cool! do you have any information you could provide on the accuracy of the software? Not trying to rag, just a stats nerd and am aware no CV model is perfect

I guess only we, the users, can provide that because what really matters is the behaviour in real life, after the model training. I played 12 games with it these last two days and it misplayed 5 times in about 300 moves. three times I just had to resign, the other two I tried to continue. Here is the type of errors I encountered

  1. I played N c6-a5, it played K b1-a1 (losing a piece: resign)
  2. I didn't play anything, it played a king move, tough game after that
    3)I didn't play anything, it played c7c6 (I needed to respond to a massive threat instead, so I resigned)
    4)played d1=Q, it played d1=R, so promotion came without check and it was very difficult to continue. I did but I lost
  3. I castled kingside with black, it played only the rook from h8 to f8 (resign)

I really appreciate this project so I'm gonna keep trying, my feeling is that these errors come mostly in the middlegame/endgame, maybe a lack of training?

The most we use it, the better it gets!!!

@computer_trash said in #20: > @BlindfoldBlunderer this is super cool! do you have any information you could provide on the accuracy of the software? Not trying to rag, just a stats nerd and am aware no CV model is perfect I guess only we, the users, can provide that because what really matters is the behaviour in real life, after the model training. I played 12 games with it these last two days and it misplayed 5 times in about 300 moves. three times I just had to resign, the other two I tried to continue. Here is the type of errors I encountered 1) I played N c6-a5, it played K b1-a1 (losing a piece: resign) 2) I didn't play anything, it played a king move, tough game after that 3)I didn't play anything, it played c7c6 (I needed to respond to a massive threat instead, so I resigned) 4)played d1=Q, it played d1=R, so promotion came without check and it was very difficult to continue. I did but I lost 5) I castled kingside with black, it played only the rook from h8 to f8 (resign) I really appreciate this project so I'm gonna keep trying, my feeling is that these errors come mostly in the middlegame/endgame, maybe a lack of training? The most we use it, the better it gets!!!
#24

@BlindfoldBlunderer said in #22:

Confusion matrix of object detector performance - imgur.com/a/Oudf6oE

Bar chart of tracker performance - imgur.com/a/ErgOCgU
(Each dataset is a set of source videos, a score of X means that X% of the video was tracked correctly before a mistake was made)

The results + people's experiences have led me to believe that if you can arrive on a "good" setup for your camera, you can get 95%+ accuracy no problem :)

Thanks, this is great!

@BlindfoldBlunderer said in #22: > Confusion matrix of object detector performance - imgur.com/a/Oudf6oE > > Bar chart of tracker performance - imgur.com/a/ErgOCgU > (Each dataset is a set of source videos, a score of X means that X% of the video was tracked correctly before a mistake was made) > > The results + people's experiences have led me to believe that if you can arrive on a "good" setup for your camera, you can get 95%+ accuracy no problem :) Thanks, this is great!
#25

@BlindfoldBlunderer: Thanks for the amazing project. Even with my very small travel set and the internal laptop camera it works like a charm. :) I am quite impressed by the performance as I started a similar projects last year with a by far worse outcome. Kudos!

Do you think it would be possible to include a "training session recording" mode that allows for adding variations to the main game? When analyzing with a board it would be nice to reiterate a position once you come up with your evaluation; or when replaying games from a book, it would be helpful to add the variations/side-lines from a given position.

I think with the current state of affairs, it is almost possible by using the feature "Start: infer {White,Black} to move", which determines the current board position. So I could play the game, go back add some point to a previous position, and re-start the recording for the side-line/variation. But I would still need to stitch everything together in the end. Would it be possible to add a button that starts a variation that is added directly to the PGN, and returns to the main game afterwards? If there is something I can contribute, please let me know (although I do not speak TypeScript yet ;-)).

@BlindfoldBlunderer: Thanks for the amazing project. Even with my very small travel set and the internal laptop camera it works like a charm. :) I am quite impressed by the performance as I started a similar projects last year with a by far worse outcome. Kudos! Do you think it would be possible to include a "training session recording" mode that allows for adding variations to the main game? When analyzing with a board it would be nice to reiterate a position once you come up with your evaluation; or when replaying games from a book, it would be helpful to add the variations/side-lines from a given position. I think with the current state of affairs, it is almost possible by using the feature "Start: infer {White,Black} to move", which determines the current board position. So I could play the game, go back add some point to a previous position, and re-start the recording for the side-line/variation. But I would still need to stitch everything together in the end. Would it be possible to add a button that starts a variation that is added directly to the PGN, and returns to the main game afterwards? If there is something I can contribute, please let me know (although I do not speak TypeScript yet ;-)).
#26

Hi,
The android app does not seem to be available anymore. Was it discontinued, or am I looking in the wrong place (google play store)?
The discord server also seems to have vanished - the invite is invalid or expired.
I as very excited by your development, but could never get it to work, maybe because it only seems to be able to get 1fps from the device camera, and I don't have an external camera to use on the laptop. I recently decided to try again, but still can't see how to get the 1fps higher.

Hi, The android app does not seem to be available anymore. Was it discontinued, or am I looking in the wrong place (google play store)? The discord server also seems to have vanished - the invite is invalid or expired. I as very excited by your development, but could never get it to work, maybe because it only seems to be able to get 1fps from the device camera, and I don't have an external camera to use on the laptop. I recently decided to try again, but still can't see how to get the 1fps higher.
#27

@b4c4d2, android app is available on google play, https://play.google.com/store/apps/details?id=com.camerachess.www.twa .
Big problem with that app is that, what for good results you need to use smartphone with very strong gpru/cpu chip.

I use Infinix Hot 40i and i have about 1-1,3 FPS, but i can use app with this FPS, but reaction time for moves takes about 5-6 seconds to happen on Lichess board. I tried to use Iphone 14 from my friend and there i have about 25-27 FPS. Very good news in that, what if you want to play on Lichess via this android app, you need only your regular chessboard and your smartphone with stand (no need pc or laptop). You can log in on Lichess on your smartphone, then launch chesscam app on smartphone and go to play chess. But bad news in that what as i said before, you need very strong smartphone to play on Lichess if you use your smartphone for that. If you use laptop/pc with your webcam, there is all good.

As i think, there is no more updates for fixing problems with FPS (i do not want buy Iphone f.e. for this).

@b4c4d2, android app is available on google play, https://play.google.com/store/apps/details?id=com.camerachess.www.twa . Big problem with that app is that, what for good results you need to use smartphone with very strong gpru/cpu chip. I use Infinix Hot 40i and i have about 1-1,3 FPS, but i can use app with this FPS, but reaction time for moves takes about 5-6 seconds to happen on Lichess board. I tried to use Iphone 14 from my friend and there i have about 25-27 FPS. Very good news in that, what if you want to play on Lichess via this android app, you need only your regular chessboard and your smartphone with stand (no need pc or laptop). You can log in on Lichess on your smartphone, then launch chesscam app on smartphone and go to play chess. But bad news in that what as i said before, you need very strong smartphone to play on Lichess if you use your smartphone for that. If you use laptop/pc with your webcam, there is all good. As i think, there is no more updates for fixing problems with FPS (i do not want buy Iphone f.e. for this).
#28

To use Chesscam you need to find right angle for view from webcam, it's not so easy. Because few moves you do, is recognized ok, but sometimes chesscam not recognizing moves.

To use Chesscam you need to find right angle for view from webcam, it's not so easy. Because few moves you do, is recognized ok, but sometimes chesscam not recognizing moves.
#29

@BlindfoldBlunderer Hello friend. I just wanted to send you this message to thank you for your ChessCam project. I am a junior programmer and I don’t know anything about Python, so I don’t understand your code very much, but I can tell how much work and effort you put into this idea.

I started playing chess one month ago and I wanted to buy a smart board because I like playing online, but I was surprised at how expensive they are. After searching a lot on Google, I found your project. I set up my classic vinyl board, my webcam and my laptop, and it feels like magic — I can play 90% or more of my games without errors.

I really don’t have words to thank you for this project. I saw that it’s been more than two months since the last update on GitHub, and I just hope you continue developing it, even slowly. Believe me, there are many people who will enjoy it.

Sorry for the long message. I wish you a great day and thank you again, my friend. You are the best.
PS: Sorry for my bad English.

@BlindfoldBlunderer Hello friend. I just wanted to send you this message to thank you for your ChessCam project. I am a junior programmer and I don’t know anything about Python, so I don’t understand your code very much, but I can tell how much work and effort you put into this idea. I started playing chess one month ago and I wanted to buy a smart board because I like playing online, but I was surprised at how expensive they are. After searching a lot on Google, I found your project. I set up my classic vinyl board, my webcam and my laptop, and it feels like magic — I can play 90% or more of my games without errors. I really don’t have words to thank you for this project. I saw that it’s been more than two months since the last update on GitHub, and I just hope you continue developing it, even slowly. Believe me, there are many people who will enjoy it. Sorry for the long message. I wish you a great day and thank you again, my friend. You are the best. PS: Sorry for my bad English.