-
## detail | 详细描述 | 詳細な説明
环境:
使用的版本是ncnn-20240102-android-vulkan
用profiler定位大致位置在stride 8, 16, 32 抽取特征时,
ex.extract("output", out);
ex.extract("771", out);
ex.extract("791", out);
![1712604972…
-
I've tried these two methods to get the ncnn model:
- The python yolov8 method:
```python
from ultralytics import YOLO
import ncnn
model=YOLO('yolov8n.pt')
model.train(data="data.yaml", ep…
-
我们自己训练的新的模型导入后无法运行,我们采用的是syenet,也是基于ncnn的,导出了param和bin文件,以同样的方式导入。
![my](https://github.com/tumuyan/RealSR-NCNN-Android/assets/167059280/f61f86b8-c030-4c72-80e1-d3d0e64be763)
-
Hi,
I am trying to deploy your project in my android phone. I modified some versions in CMake file and the project can be successfully built. Then I compiled the ncnn library. As the newer version …
-
## error log | 日志或报错信息 | ログ
**出错代码:
ex.extract(one_param_id::BLOB_images, in_pad);**
出错信息
A/libc: Fatal signal 11 (SIGSEGV), code 1, fault addr 0x0 in tid 25647 (ImageReader-192)
##…
-
1、检测框没问题,但是识别的结果都是单字的, 并且不对
- ncnn版本ncnn-20220420-android-vulkan
- Opencv版本4.5.4
-
你好大佬,我按照你发的https://github.com/Digital2Slave/ncnn-android-yolov8-seg/wiki/Convert-yolov8%E2%80%90seg-to-ncnn-model-step-by-step 转换了一个模型,就是部署上去闪退,找不到原因。
-
感谢大佬提供的移动端opencv库,本人需要用到人脸检测矫正函数,所以需要添加calib3d模块,但是calib3d默认使用了rtti选项,这里通过一个死方法,打开了rtti选项才将calib3d模块成功添加后才编译通过,模型有点大;ncnn也必须开启rtti,才能同时使用ncnn和opencv。这里是具体的解决办法 [开启rtti选项源码编译android的ncnn库](https://blog…
-
@nihui Thanks for your sharing. I converted the `torchvision.models.resnext101_32x8d`to `ONNX` and then to `NCNN`. This process did not report any errors,but I used the NCNN model to get extremely lar…
-
Build command failed.
Error while executing process C:\Users\xbo\AppData\Local\Android\Sdk\cmake\3.10.2.4988404\bin\ninja.exe with arguments {-C F:\workspace\as_work\ncnn-android-yolov5\app\.cxx\cmak…