MobilenetSSD_caffe

How to train and verify mobilenet by using voc pascal data in caffe ssd?

Introduction

• Help to detect objects by using trained MobileNet SSD models and detect script.
• I downloaded PASCAL-VOC2007/VOC2012 data, and trained and tested MobileNet SSD models in caffe ssd.
• I developed android jni libs of MoblieNet SSD, and compiled question which you can read my caffe_ssd-android-lib
• It can detect 20 classes objects included animal/traffic/house/people etc.

Requirements

• Caffe SSD
• Android Studio
• Python

Usage

• Get code

git clone --recursive https://github.com/avBuffer/MobilenetSSD_caffe.git
cd MobilenetSSD_caffe

• Train and test MobilenetSSD models, in folder of mobilenet_caffe_train

  • Step1: Download and create lmdb of PASCAL-VOC2007/VOC2012, and then put train and test lmdb into data/VOC0712 folder
  • Step2: Modify your caffe ssd path in files of train.sh/test.sh and demo.py/merge_bn.py in tool folder and other folders
  • Step3: Do train and test, and get mobilenetssd iter caffemodel

    cd MobilenetSSD_caffe/mobilenet_caffe_train
    ./tool/train.sh
    

  • Step4: Change mobilenetssd iter caffemodel

    cd MobilenetSSD_caffe/mobilenet_caffe_train
    ./tool/merge_bn.py
    

  • Step5: If you want detect tests, you can do next steps

    cd MobilenetSSD_caffe/mobilenet_caffe_train
    ./tool/demo.py
    

• Detect and classify objects by Mobilenet SSD model in PC in the folder of classify_caffe, you can do next steps

  • Step1: Modify your caffe ssd include and libs path in CMakeLists.txt
  • Step2: Compiler

    cd MobilenetSSD_caffe/classify_caffe
    ./build.sh
    

  • Step3: Classifier

    cd MobilenetSSD_caffe/classify_caffe
    ./classify.sh
    

• Detect and classify objects by Mobilenet SSD model in Android in the folder of MobilenetSSD_APP_demo, you can do next steps

  • Step1: Import MobilenetSSD_APP_demo project into Android Studio
  • Step2: Build APK in Android Studio, and install APK into Smartphone
  • Step3: Create caffe/mobilenet folder in Smartphone SDCard

    cd MobilenetSSD_caffe/classify_caffe/model/mobile/
    adb shell mkdir caffe
    adb shell mkdir caffe/mobilenet
    adb push MobileNetSSD_deploy.caffemodel sdcard/caffe/mobilenet
    adb push MobileNetSSD_deploy.prototxt sdcard/caffe/mobilenet
    

  • Step4: You can run APP to select picture or take picture to detect objects

Issues

• If you have any idea or issues, please keep me informed.
• My Email: jalymo at 126.com, and my QQ/Wechat: 345238818

Wechat&QQ group

• I setup VoAI Wechat group, which discusses AI/DL/ML/NLP.

• VoAI means Voice of AI, Vision of AI, Visualization of AI etc.

• You can joint VoAI Wechat group by scanning QR-code in path ./imgs/VoAI.jpg.

• 

• Also you can joint QQ group ID: 183669028

Any comments or issues are also welcomed.Thanks!