wms2537

But the model does not seem to converge after a few epochs. The mAp stays at around 0.001

May I know what is the initial loss of your model? Or could you post your training plot ?

Problem still exist when building on Jetson NX ```[ 97%] Building CUDA object CMakeFiles/mxnet.dir/src/operator/tensor/la_op.cu.o [ 97%] Building CUDA object CMakeFiles/mxnet.dir/src/operator/tensor/matrix_op.cu.o [ 97%] Building CUDA object CMakeFiles/mxnet.dir/src/operator/tensor/ordering_op.cu.o [ 97%] Building CUDA...

Isn't it turned on by default, I used the code pulled from master, the problem still exists. I can compile it on normal pc but not on jetson.

Here's the cmake output: ```-- The C compiler identification is GNU 7.5.0 -- The CXX compiler identification is GNU 7.5.0 -- Check for working C compiler: /usr/bin/cc -- Check for...

Still the same: ``` io.cc:(.text+0xa8): relocation truncated to fit: R_AARCH64_CALL26 against symbol `std::__cxx11::basic_string::_M_create(unsigned long&, unsigned long)@@GLIBCXX_3.4.21' defined in .text section in /usr/lib/gcc/aarch64-linux-gnu/7/libstdc++.so io.cc:(.text+0xc0): relocation truncated to fit: R_AARCH64_CALL26 against symbol...

I think my system toolchain is up to date, I am using jetpack 4.3. If not, how to update system toolchain?

After some testing, I finally managed to build it. I updated ccache and openblas similar to > Please ensure your system toolchain is up to date (includes https://bugzilla.redhat.com/show_bug.cgi?id=1243559 fix) >...

Which version of esp-idf are you using? Try to use something above 5.0.

The latest is `5.1.1` [https://github.com/espressif/esp-idf/releases/tag/v5.1.1](https://github.com/espressif/esp-idf/releases/tag/v5.1.1)