JuZiSYJ / MobilenetSSD_Android

using ncnn, on android, detection, model size only 11MB, 10fps on 845/cpu

MobilenetSSD_Android

using ncnn, on android,detection,surport arm v7/v8/x86, surport vulkan on gpu

environment

ncnn
AndroidStudio3.2(NDK19)
Qualcomm845

model

using pretrained MobilenetSSD on VOC，bychuanqi305,mAP=0.72
notice：the model in ncnn/example/mobilenetssd is too old to perform，please use ncnn-assert
pipeline: model on PC --> ONNX ---> model on ncnn(.bin,,param) --> inference on Android
converted model on ncnn is from ncnn-assert

organization

using ncnn and opencv on android
the sim-overall android-organization is here

result

infer time on 845 in float32 is aroud 100ms, depending on the object

ncnn is good in optimazing on float32

apk

the apk is available on BAaiduYun,password is 3alh

reference

(https://github.com/nihui/ncnn-android-squeezenet)
(https://blog.csdn.net/qq_33431368/article/details/84990390)