optical-breacher
optical-breacher copied to clipboard
govizlora
Add ability to manually enter/modify sequences
Sometimes the sequences on the right side are not OCR'd correctly, either missing a row or wrong characters. I would like to be able to add/edit rows manually after they are scanned.
I don't want to raise another issue for this, but is there any way community could help you with accuracy of this app? I mean app frequently confuse BD with 55 or vice versa. Editing results would be great. What do you think if we provide you some screenshots so you could train this app to be more accurate?
Bonus - let upload already taken pictures and cut them. This way pictures could have higher quality and be recognized better, because as of now they are taken a bit blurry because they are very close to screen in my case. I believe blur is causing the majority of false recognitions.
Thank you all for the help! I've recently retrained the model, and hopefully the accuracy is improved: https://github.com/govizlora/optical-breacher/pull/21
At https://govizlora.github.io/optical-breacher/ you can find the version at the bottom left corner, if it >= v1.7, it is using the new model.
Adding the ability to modify is a smart idea and I'll try to add it!
In the long run, I want to try to replace the current approach with tesseract entirely with OpenCV + CNN, but there is no guarantee that it will happen :p Community help with this approach would be really appreciated.
Btw, uploading the photo from your phone should be possible now, you can find the SCREENSHOT MODE at the bottom right corner.
I don't want to raise another issue for this, but is there any way community could help you with accuracy of this app? I mean app frequently confuse BD with 55 or vice versa. Editing results would be great. What do you think if we provide you some screenshots so you could train this app to be more accurate?
Bonus - let upload already taken pictures and cut them. This way pictures could have higher quality and be recognized better, because as of now they are taken a bit blurry because they are very close to screen in my case. I believe blur is causing the majority of false recognitions.
That would be great sorry just got back from Afghanistan give me a few days I’ll get onto it
