Cloud Text Recognition don't work
kobidy1102 opened this issue · comments
I get an "urlList is empty, failed to detect cloud text" error when I try to recognize the text from the Album right in your sample code.
Please fix it
https://www.youtube.com/watch?v=aU4Y1pKiDTo
This is a video of the bug. (I run code module-text)
It has an error with any new photos or photos from the album.
I use a Xiaomi Mi9T, android 10.
Hope you can fix it so I can soon put this SDK into my app.
Hi，you can check the README.md of "MLKit-Sample",
https://github.com/HMS-MLKit/HUAWEI-HMS-MLKit-Sample/tree/master/MLKit-Sample
I pasted the part below， and for cloud api has a certain free quota, if it exceeds the free quota, you need to apply separately。
And are you willing to tell us the name of the app, we can have a relevant experience：）
If you want to use cloud analyzer, such as cloud text analyzer(document and cloud text), cloud image classification analyzer, landmark analyzer, translation, language detection, you need to apply for an agconnect-services.json file in the developer alliance(https://developer.huawei.com/consumer/en/doc/development/HMS-Guides/ml-preparations-4), replacing the sample-agconnect-services.json in the project.
Attention: The package name in this project can not be used to apply for agconnect-services.json. You can use a custom package name to apply for agconnect-services.json.
You only need to modify the applicationId in app HUAWEI-HMS-MLKit-Sample\app\build.gradle to the same package name as the applied agconnect-services.json to experience the mlkit on-Cloud services.
Before I created this issue, I tried your SDK. I created a project at AppGallery Connect. I downloaded the json file and tried it in the project, but got the error above so I downloaded the sample project to see and it gave the same error.
After reviewing, I realized I was missing a step of "apply plugin: 'com.huawei.agconnect'". It works now, but the document on the site is quite rudimentary, and the segment is not designed to guide developers through it step by step, I think many other developers will have trouble. Get started with your guide soon.
I have an application with more than 1 million downloads but before putting the SDK into my application, I will try it on another application.
How many free quota are used in a month? When the current limit is exceeded, will the Text Recognize Cloud work? I hope the price of the service will be much better than Google Cloud.
Hi, thanks for you feedback , we will consider to optimize our document
and for the free quota question , can refer to below table :
since now ,the charging strategy has not come out. If you want to increase the free quota, you can apply by email and tell us your app information and the free quota you expect. We will evaluate and adjust it according to your APP.
Quota adjustment method: Sign in to AppGallery Connect, go to Develop > Growing > ML Kit, and query the number of on-cloud API calls. If you need to change the number, download the email template, fill in the template as instructed, and send it to mlkit@huawei.com.
| Category | Service | Device | Free-of-Charge Quota (Device) | Cloud | Free-of-Charge Quota (Cloud) |
|---|---|---|---|---|---|
| Text services | Text Recognition | √ | Unlimited | √ | 10,000 calls/month |
| Text services | Document Recognition | N/A | √ | 10,000 calls/month | |
| Text services | Bank Card Recognition | √ | Unlimited | N/A | |
| Text services | General Card Recognition | √ | Unlimited | N/A |
Thanks for your information.
I am currently using Google Cloud Text Recognition. After a period of testing I have a few feedbacks as follows:
- Advantages:
- Fast speed
- Easily switch from Google Cloud to your SDK without having to fix too much logic code
- Accuracy for high Latin letters
- Limitations:
- The accuracy of languages not using Latin letters should be improved. I predict Google Cloud has an accuracy of about 99%, and your SDK about 90%. This is not a bad rate, but it needs to be improved to 95% + (A sentence with 10 words but only need to identify one word wrong will make the meaning of the sentence change).
- I'm really dissatisfied with dense text recognition (OCR_COMPACT_SCENE). 2 rows or columns are very far apart but the result is still about the same Block. Because of this problem, I haven't been able to put it into my application yet. I hope your team improves it.
Thank you very much for your feedback. Your feedback is very valuable for our product improvement. Two questions about you:
- The accuracy of languages not using Latin letters
---- Can you tell us what kind of non Latin letters text you are testing? We will reproduce the problem and make improvements. - For the second problem, we will analyze and confirm as soon as possible, if you can give us one simple picture , it will be best .
This is a picture I have tested, it has lots of misidentified words. And the dense text recognition I don't know on which factors you segment into blocks, I think it's necessary to calculate the distance between discrete blocks (maybe also based on the text color) to combine it into a Block. Hope to be helpful to your team developing this SDK.
Hi, since now, the cloud api don't support Traditional Chinese, only support Simplified Chinese.
on-cloud API supports much more languages such as Simplified Chinese, English, Spanish, Portuguese, Italian, German, French, Russian, Japanese, Korean, Polish, Finnish, Norwegian, Swedish, Danish, Turkish, Thai, Arabic, and Hindi.
This is a picture I have tested, it has lots of misidentified words. And the dense text recognition I don't know on which factors you segment into blocks, I think it's necessary to calculate the distance between discrete blocks (maybe also based on the text color) to combine it into a Block. Hope to be helpful to your team developing this SDK.
Hello, regarding the problem you feedback last time, we have located the problem and fixed it. We have synchronized the repaired version to Singapore and China sites. Other global sites such as Europe will also be completed this week， you can try the current effect. and for the traditional Chinese support , we are trying to build this capability， When we have a support plan, I will inform you in time, thank you again for your feedback ，hope you can continue to provide us with more valuable questions and suggestions
Befor the issue fixed:
After the issue fixed:
Sorry for being able to answer now, I'm quite busy with current projects.
Thanks for fixing the "dense text recognition", I will test it.
Hopefully the SDK will support traditional Chinese soon.
I have heard you talk about China sites, Singapore sites, Euro sites, .. Are these servers ?, and if my users are in Euro but I use Singapore sites then Cloud will respond more slowly when using Euro sites, right?