const float mean_vals[3] = {104.f, 117.f, 123.f};
in.substract_mean_normalize(mean_vals, 0);
ncnn::Extractor ex = squeezenet.create_extractor();
ex.set_light_mode(true);
ex.set_num_threads(4);
ex.input(squeezenet_v1_1_param_id::BLOB_data, in);
ncnn::Mat out;
ex.extract(squeezenet_v1_1_param_id::BLOB_prob, out);
__android_log_print(ANDROID_LOG_DEBUG, "SqueezeNcnn", "Output Channels : %d", out.c);
cls_scores.resize(out.c);
for (int j=0; j<out.c; j++)
{
const float* prob = out.channel(j);
cls_scores[j] = prob[0];
//cls_-scores[j] = out[j];
}The above code is implemented in Android Studio. I've printed out parameter "in", which is the correct image value, but output channel(out.c) = 0 and cls_scores = 0. The libncnn.a is compiled through
cmake -DCMAKE_TOOLCHAIN_FILE=$ANDROID_NDK/build/cmake/android.toolchain.cmake
-DANDROID_ABI="arm64-v8a" -DANDROID_PLATFORM=android-21 -DCMAKE_INSTALL_PREFIX=/home/liuwen/ncnn_lib ..
Would anyone please give suggestions how this could happen, I've been stucking into this problem for a while. Thanks!
2条答案
按热度按时间gfttwv5a1#
int inSucess = ex.input(squeezenet_v1_1_param_id::BLOB_data, in); __android_log_print(ANDROID_LOG_DEBUG, "SqueezeNcnn", "In Success : %d", inSucess);
2020-01-07 17:53:39.122 27269-27269/com.example.hellojni D/SqueezeNcnn: In Success : -1
After checking the correctness of ncnn::Mat, I'm not sure why loading ex.input the result is incorrect. (return 0 is success ; return -1 is not success)
zd287kbt2#
This usually means that the model loading has already failed.
If you use xyz.param path, use Net::load_param(const char*) to load from param memory
If you use xyz.param memory, use Net::load_param_men(const char*) to load from param memory
If you use xyz.param.bin path, use Net::load_param_bin(const char*) to load from param memory
If you use xyz.param.bin memory, use Net::load_param(const unsigned char*) to load from param memory