The performance of libcaffe.so/libcaffe_jni.so of armeabi-v7a is slower than that of arm64-v8a

[hgffly]

Hi sh1r0,
I have tried caffe-android-demo with my trained model. For recognizing an image, it's take about 1.5s.
But due to the requirement of integrating caffe with my developed app, I need to use libcaffe.so/libcaffe_jni.so of armeabi-v7a. I built this version of libraries with settting export ANDROID_ABI="armeabi-v7a-hard-softfp with NEON" and NDK r11c .
To evaluate the performance, I replace caffe.so/caffe_jni.so of arm64-v8a in caffe-android-demo with the ones I built. I just found that it takes about 4.8s to recognize an image, about 3 times slower.
Is it correctly? Did I miss something? Is it possible to correct or improve it ? Thanks