equationl
equationl
Thanks for remind, Actually you can use `1.2.8` replace `v1.2.8` . `1.2.8` is build successful, I can't find out why jitpack auto build 1.2.8 instead of v1.2.8 .
@cagnulein As I said in README, I don't push model file to fastdeploydemo, because it too big, so you need download model to fastdeploydemo by yourself. > 因为 Paddle 模型比较大，所以没有集成模型到...
It's seem ndk version problem, try about this: https://github.com/android/ndk/issues/329
Sure, I will test it, But I don't have android 14 device, So I just test in simulator
@AppWerft Yes, it may be banned by google play, but I don’t know why my app uses this library, but it can also be uploaded to google play.
What does Google say?
Ok, Get it.
@HBiSoft Yes, I see. I am thinking how to solve it
你下官方的 demo 替换成这张图片看看能不能复现出这个问题：https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.7/deploy/fastdeploy/android 因为理论上来说，我这个项目只是二次封装了上述 demo ，没有更改其他任何参数配置，识别问题一般是原本项目的问题，但是也不排除可能是我封装时某些地方写的有问题，所以需要你使用上述原 demo 确认一下。
就是官方的 v3 slim量化 模型，这个识别效果还却决于运行的设备。 同样的模型和demo，我自测使用 macos 的 arm 模拟器识别效果最佳（就是 README 演示的效果），但是使用真机（mi 10u）效果就不是太理想，就如你所说，甚至检测模型都无法检测正确的文本分段。