Paddleocr

[cagnulein]

No description provided.

[cagnulein]

@victorypoint it's ready for you test when you will come back (i should probably modify the return of the paddle engine because it's returning some verbose things so probably qz will not handle them yet, but the biggest part has done).
Have a nice trip!

[victorypoint]

@cagnulein, I'm back from Scotland and just tested this latest APK. I made a number of speed and incline changes. Here is the logcat:

stream-logcat.txt

[cagnulein]

welcome back @victorypoint ! I hope you had a great vacation in Scotland!
Thanks for the test, it indeed worked but only with words, i can't see any number. The problem could be 2: the chinese assset that i'm using (but the number should be the same in chinese) or the reduced image that i was using to improve the performance.

I built a new one with the full image https://github.com/cagnulein/QZCompanionNordictrackTreadmill/actions/runs/10678955998/artifacts/1884857824

Let's collect another log from this one, let's see if it's better. I saw in your previous one that the time that paddle took for the small image was about 1 second, not bad but also not perfect

[victorypoint]

@cagnulein, I just tested the new apk and have attached the logcat. Again I made a number of speed and incline changes.
stream-logcat.txt

[cagnulein]

thanks @victorypoint there is some issue that i have to address that i didn't see on my android simulator. it seems it can't process numbers at all. that's very strange. maybe because it's a different ARM type? maybe it's a 32bit one?

[victorypoint]

Ok no worries. I've been using your last APK that uses Google OCR as it's working fine for me with iFit Beta regardless of the OCR delay. The only problem is I have to do some fiddling around to get the APK to show the "start ocr recording" prompt. Could this be modified so ocr recording is on at startup?

Also, is it worth doing a build that uses AI.Server for OCR much like what you've done with QZ?

[cagnulein]

Do you mean that you can't see the prompt request all the time? For the ai server : I have to check it. We can use a udp socket to send it? Roberto Viola Software engineer and open source enthusiast http://robertoviola.cloud Il giorno ven 6 set 2024 alle 17:57 Al Udell ***@***.***> ha scritto:

…

Ok no worries. I've been using your last APK that uses Google OCR as it's working fine for me with iFit Beta regardless of the OCR delay. The only problem is I have to do some fiddling around to get the APK to show the "start ocr recording" prompt. Could this be modified so ocr recording is on at startup? Also, is it worth doing a build that used AI.Server for OCR much like what you've done with QZ? — Reply to this email directly, view it on GitHub <#108 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAALYWEAVE22FXRYM2AUC73ZVHGHHAVCNFSM6AAAAABNH4NVFCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMZUGM3DINBQHA> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[victorypoint]

Right, the prompt doesn't appear on TM boot. I have to stop and start iFit and Companion a few times to get the prompt to appear.

For Ai.server, we're sending a request to an OCR server like: ocr = requests.post("http://localhost:32168/v1/image/ocr", files={"image":image_data}).json(). The 3 python scripts in QZ handled the cropping and image processing of Zwift screens. So something like that in Companion. UDP may work?

[cagnulein]

Ok I will check for the prompt. For ai server we will need to specify also an ip on the companion Roberto Viola Software engineer and open source enthusiast http://robertoviola.cloud Il giorno ven 6 set 2024 alle 19:04 Al Udell ***@***.***> ha scritto:

…

Right, the prompt doesn't appear on TM boot. I have to stop and start iFit and Companion a few times to get the prompt to appear. For Ai.server, we're sending a request to an OCR server like: ocr = requests.post("http://localhost:32168/v1/image/ocr", files={"image":image_data}).json(). The 3 python scripts in QZ handled the cropping and image processing of Zwift screens. So something like that Companion. UDP may work? — Reply to this email directly, view it on GitHub <#108 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAALYWBI3QYRJEJUXDIO753ZVHOC7AVCNFSM6AAAAABNH4NVFCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMZUGQ3TOOJWGI> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[victorypoint]

Let's use 192.168.1.4 for ai.server IP.

[cagnulein]

this is for me for reference

[cagnulein]

Right, the prompt doesn't appear on TM boot. I have to stop and start iFit and Companion a few times to get the prompt to appear.

For Ai.server, we're sending a request to an OCR server like: ocr = requests.post("http://localhost:32168/v1/image/ocr", files={"image":image_data}).json(). The 3 python scripts in QZ handled the cropping and image processing of Zwift screens. So something like that in Companion. UDP may work?

i will use this as a prompt for cursor ai editor, i would like to see its functionality :)

[cagnulein]

@victorypoint i found the root issue of the paddleocr built, it can't recognize the small fonts on the incline and speed metrics on the bottom. there is any way to make them bigger in the ui settings? otherwise can you try on paddle on pc if you can get it?

in the meantime i'm adding the ai server support

[victorypoint]

@victorypoint i found the root issue of the paddleocr built, it can't recognize the small fonts on the incline and speed metrics on the bottom. there is any way to make them bigger in the ui settings? otherwise can you try on paddle on pc if you can get it?

in the meantime i'm adding the ai server support

@cagnulein, no sorry, there appears to be no way currently to resize UI elements in iFit beta. I'll keep investigating.

[cagnulein]

ok let me know if you can play with paddle on windows on it. thanks! Roberto Viola Software engineer and open source enthusiast http://robertoviola.cloud Il giorno ven 13 set 2024 alle ore 15:40 Al Udell ***@***.***> ha scritto:

…

@victorypoint <https://github.com/victorypoint> i found the root issue of the paddleocr built, it can't recognize the small fonts on the incline and speed metrics on the bottom. there is any way to make them bigger in the ui settings? otherwise can you try on paddle on pc if you can get it? in the meantime i'm adding the ai server support @cagnulein <https://github.com/cagnulein>, no sorry, there appears to be no way currently to resize UI elements in iFit beta. I'll keep investigating. — Reply to this email directly, view it on GitHub <#108 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAALYWBAJYVYCGYNNDLVFXDZWLTNDAVCNFSM6AAAAABNH4NVFCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNBYHE4TANBZGI> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[victorypoint]

@cagnulein, yes absolutely we can try paddleocr windows on the ifit images. How would I get the images your generating in companion?

[cagnulein]

i guess you can use the screenshot

it's what i'm using to test it

[victorypoint]

@cagnulein, I ran both PaddleOCR Windows and ai.server against the above screenshot. Both versions convert all the screen elements to OCR with very little error. Note that the raw OCR output in the attached logs is formatted a bit different.

ai.server-log.txt
paddleocr-log.txt

[cagnulein]

Perfect, there is hope, probably i need to use the English model instead of the Chinese one Roberto Viola Software engineer and open source enthusiast http://robertoviola.cloud Il giorno ven 13 set 2024 alle 17:15 Al Udell ***@***.***> ha scritto:

…

@cagnulein <https://github.com/cagnulein>, I ran both PaddleOCR Windows and ai.server against the above screenshot. Both versions convert all the screen elements to OCR with very little error. Note that the raw OCR output in the attached logs is formatted a bit different. ai.server-log.txt <https://github.com/user-attachments/files/16995947/ai.server-log.txt> paddleocr-log.txt <https://github.com/user-attachments/files/16995950/paddleocr-log.txt> — Reply to this email directly, view it on GitHub <#108 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAALYWEVPH4DTSXGXWVCA5TZWL6RTAVCNFSM6AAAAABNH4NVFCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNBZGE4TMMRTGA> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[cagnulein]

i tried also the paddleocr4android app as it is with our screenshot and it doesn't work either. i'm trying to issue a new on their repository

[cagnulein]

ticket here equationl/paddleocr4android#42

[cagnulein]

hi @victorypoint are you aware of this equationl/paddleocr4android#42 (comment) ?
I'm trying to port this to Fastdeploy then

[victorypoint]

@cagnulein, looks like good news on that ticket. Very helpful fellow. Since I'm using the Windows python deploy, it uses the PP-OCRv4 English model by default. I haven't tried the other models but read they are very good as well. https://github.com/PaddlePaddle/PaddleOCR/blob/main/doc/doc_en/quickstart_en.md

[cagnulein]

Yes with the model proposed by the mantainer I was already able to see speed and incline on the demo project. I will port it to this branch.
THe only thing that I saw it's that on the emulator it's very slow, like 3 seconds per image, so i guess on your tablet could be only slower.
But let's see when I will complete the port

[cagnulein]

done https://github.com/cagnulein/QZCompanionNordictrackTreadmill/actions/runs/15022482902

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.