out.h w c are all zero after replacing my own .so file

[YiLin-Chen]

I found that my matrix out is empty
ncnn::Mat out;
ex.extract(squeezenet_v1_1_param_id::BLOB_prob, out);

I have read #391 #248 #396, and none of their solution can solve my problem. Below is my building steps,

1. build .a file for android according to the article https://github.com/Tencent/ncnn/wiki/cmake-%E6%89%93%E5%8C%85-android-sdk

2. modify Android.mk file in ncnn-mobile squeezencnn-AS app (https://github.com/dangbo/ncnn-mobile), just change NCNN_INSTALL_PATH := /my/path/to/ncnn/build-android/install

3. modify Application.mk file by commend NDK_TOOLCHAIN_VERSION := 4.9 (same as 为什么我拿ncnn编译出来的.a文件编译so的时候会报undefined reference to '__kmpc_fork_call'的错误 #292 )

4. generate libsqueezencnn.so by ndk-build

During the above process, I didn't get any error message, however, the result is wrong. The output is alway "tench, Tinca, tinca = 0.000" and out.c out.w out.h are all zero. I use the original model and code as ncnn-mobile squeezencnn-AS provides.

I use Android Studio 3.1.2 on macOS High Sierra with ndk-r16b. Actually the prebuild .a file is not working for me either.

Thanks.

[YiLin-Chen]

I found that the return value of load_param and load_model are 0. so weird... I didn't change the model.

[YiLin-Chen]

After digging into the source code, I think it runs into

if (magic != 7767517)
{
fprintf(stderr, "param is too old, please regenerate\n");
return 0;
}

[nihui]

the model bundled in squeezencnn-AS project is too old to work with the latest ncnn libraray. try replacing model files with the latest version in ncnn repo

[YiLin-Chen]

Thanks for your reply. I change the whole project to ncnn/examples/squeezencnn and use both prebuild and my own .a files, however, the result is same as before. ret = 0 from load_param and load_ model

[YiLin-Chen]

Okay, I solve the problem and you are right. The model in ncnn-mobile repo is too old to be used. Please forget my last comment, I might have something wrong with my operations. To conclude, I solved the problem after I replace the ncnn-mobile project with the project in ncnn/examples/squeezencnn. Thanks :)