Poker-AI.org

I am new to botting and probably the lowest coding/computer IQ person on this forum but know poker very well. I saw a PokerACR that can push data to PokerSnowie and ICMizer and take action after that instantly. I am looking for a similar set up that will work on WSOP or Americas Cardroom.

Is there one out there?

My name is Dmitry. I am a project manager PokerOCR.
Our development - PokerOCR(PokerRecorder + PokerRecorderOCR) + BotPS.
Creates hand histories "on the fly". Uses only OCR. Works either through RDP or VirtualBox. Works together with solvers: PokerSnowie and ICMIZER. The security service of poker rooms has no chance to determine the presence of this assistant during the game.
Now we are implementing support for 888. Now the software supports two rooms: PokerStars, TitanPoker.

If there is interest in our project, then, if the administration of the poker-ai.org resolves, it will be possible to create a topic where we will answer your questions. In our experience, there are always a lot of questions, especially with installation and configuration.

Official site: http://www.pokerocr.com/en/
Support: [email protected]

Let me know if you want a sub forum for this

how much time it needs to read table information for 1 frame? What winrates does your bot have considering you are using ICMIZER and
snowie

Sorry for my english.
From the moment the Hero makes his turn until the creation of the Hand History (9 max) for 2-3 seconds. Next is sending to PokerSnowie/ICMIZER. PokerSnowie/ICMIZER processes the Hand History in 5-6 seconds. Total from the moment of the Hero’s turn to pressing the button no more than 7-9 seconds.


Tested at different limits. Winrate not considered. Playing with PokerSnowie - is a small plus (apart from rake).

Below are PokerOCR video games using PokerSnowie:

video#1
video#2

Thank you for the reply your english is good.

8 seconds per move seems to me to be very obvious tell that it is a bot, humans never take so long for button click for every single move.
Anyway i am more interested in your OCR part, as it is time consuming and boring task for most developers, and so many people need it. Do you plan soon to make it for new poker sites like betonline, ACR etc, and how long would it take you to adjust just OCR part for a new site? I would be interested in buying your OCR solutions, but not interested in pokerstars as it is high security and no rake back, tough player pool. I am interested in ipoker, how much would you request for one time payment for OCR part for ipoker? and could you deliever solution in the form of dll so that i can integrate it in my project?
For OCR i currently use openholdem tablemaps, but the issue is that also tablemaps you need to do yourselves, take care of updates etc.

Regarding your mouse movements, at some points in video it looks natural movement (as by human) and at some point looks more as mouse teleport. Do you use some human mouse recordings?

The program has an autofolder. Autofolder works in 60-80% of the hands. Autofolder determines the presence of Fold immediately after creating a hand history (2-3 seconds).
The user can press the Fold button randomly on time in diagonal zone 2-9 seconds. Thereby veiling the presence of the bot. Players sometimes think for 40 sec.


Do you mean the recognition of stacks, cards, pots, bets or the names of players?


We are currently working on 888, further PartyPoker


If only OCR is about 2 weeks


We do not provide software development services.


Auto-click is made in the random area of ​​the button. You can always turn off auto-click.
The program also supports hotkeys.

Our software is an intermediary between the solver and the poker table. Solver board can be ignored and played by yourself.

Yes i am interested on all of it. But if you don't offer it as a DLL then there is not much benefit in purchasing OCR part.

PokerOCR consists of two programs:
PokerRecorder - takes screenshots and creates hand histories;
PokerRecorderOCR - OCR program;
If you are interested, you can create an API for accessing third-party programs to PokerRecorderOCR.
The formation of the image will lie on the program that will use the API.

Ok sounds like a good idea, i would wait with that until you enable more poker sites. Does your OCR works only for specific table size and design? If pokersite would just a bit change design / fonts would it still work?

I agree


The neural network is trained on a variety of images. This allows in small limits to change the size of the table and font.

Ok i guess the more data you collect for fonts, easier it gets to recognize them on new sites and maybe even it would work at some point on any new site without fine tuning neural network? Especially if you would use different sizes of table. It should be easy to collect data.

I guess you use some of CNN models?

We create our own neural network for each site. We use convolutional neural network.

Why do you use new neural network for each site? I think that way neural network overfits to that poker site, which would be the same as template matching technique that most of botters use.
I think the only point of using neural network would be to train it with a lot of images of different scales of fonts and different sites so that it generalizes well on new unseen sites or when poker site updates fonts.

The structure of the neural network, we have one for all types (stack, pot, board, etc.). Neural networks are different.
If the size and style of the font, for example, the stack on a new site are about the same, then we take a neural network from another site and add new data to it. If the differences are serious, for example, board maps, then we create a new neural network and train it.
You can create one neural network, for example for a board, and train it with new data from new sites. But this may entail false network responses. And it will have to be taught every time longer and longer. The average time to create hand histories with our software is ~ 1500-1900 ms.
Screen PokerRecoder

I still don't see a need to use new neural network even if fonts or cards images look quite different.
you just need to provide enough data to neural network so that it can generalize well to new unseen fonts.

Do you use data augmentation?

If the size and style of the font, for example, the stack on a new site, are about the same, then we take a neural network from another site and add new data to it.

By data augmentation i meant if you increase your training samples by making image transformation on the original input samples, like adding some noise, scaling it etc. That way you would
get a lot more data and better generalisation. Do you use some preprocessing on input samples before you train the network?

Yes, we use reasonably scale, offset

If you find,contact pm me pls