Created on 2017.07.10

• Requirement
  • OS : Ubuntu
  • Dependency
    • python2.7++
    • openCV 2.4.8++
    • CUDA 7.5++
    • cudnn4++
    • caffe
  • Dataset
    • GTSRB Homepage
    • Dataset Download
      • train dataset
      • test dataset
      • test gt dataset
        

• Directory

  • images
    • README.md에 삽입되는 image 및 기타 image저장
  • label
    • GTSRB dataset(test/training)에 대한 label 파일 저장
  • lenet
    • leNet network prototxt 저장
  • resNet
    • resNet(20/32) network prototxt 저장
  • old_script
    • previous lmdb creating script

• Script file

  • build.sh
    • caffe build script
  • create_data_list.py
    • creating dataset(GTSRB) label script
  • get_data.py
    • download dataset(GTSRB) script
  • create_lmdb.py
    • create dataset(GTSRB) lmdb & mean_binaryfile script
  • test_lenet.sh
  • test_resnet_20.sh
  • test_resnet_32.sh
  • train_lenet.sh
  • train_resnet_20.sh
  • train_resnet_32.sh  

1. Caffe build

* Using Script
  ./build.sh [CAFFE_HOME_PATH]

* Using Bash shell
 cd $CAFFE_HOME
 make -j[cpu core] --> make -j8

2. Download GTSRB dataset

• Default dataset store : $HOME/data/GTSRB
  • Training data : $HOME/data/GTSRB/Final_Training/Images
  • Test data : $HOME/data/GTSRB/Final_Test/Images
  • Test_gt_csv file : $HOME/data/GTSRB/GT-final_test.csv

* Using Script
 cd $CAFFE_HOME/examples/traffic_sign_recognition
 python get_data.py

3. Create data label

• Label Format
  • Training

  [class folder name]/[image filename] [class id]
  • Test

  [image filename] [class id]
• Label file store
  • Training
    • $HOME/data/GTSRB/Final_Training/Images/train_label.txt
    • $CAFFE_HOME/examples/traffic_sign_recognition/label/train_label.txt
  • Test
    • $HOME/data/GTSRB/Final_Test/Images/test_label/txt
    • $CAFFE_HOME/examples/traffic_sign_recognition/label/test_label.txt

* Using Script
 cd $CAFFE_HOME/examples/traffic_sign_recognition
 python create_data_list.py

4. Create LMDB

• LMDB store path

- default path : $HOME/data/GTSRB/lmdb/[train_lmdb/test_lmdb]
- link : $CAFFE_HOME/exmaples/traffic_recognition/lmdb/[train_lmdb/test_lmdb]

• LMDB store path

- default path : $HOME/data/GTSRB/lmdb/[train_mean.binaryproto]
- link : $CAFFE_HOME/exmaples/traffic_recognition/lmdb/[train_mean.binaryproto]

* Using Script
python create_lmdb.py

usage: create_lmdb.py [-h] caffe_root

positional arguments:
 caffe_root  Please Input CAFFE HOME PATH(eg. /home/user/caffe) 

5. Training

• example network model
  • lenet : /models/lenet
  • resnet : /models/resnet20 && /models/resnet32
• default lmdb path
  • train : /lmdb/train_lmdb
  • test : /lmdb/test_lmdb
  • mean_binary : /lmdb/train_mean/binaryproto

* Using Script
cd $CAFFE_HOME/examples/traffic_sign_recognition
python train.py

usage: train.py [-h] [--gpus GPUS] [--run-soon]
              [--network-model-name NETWORK_MODEL_NAME]
              caffe_root

positional arguments:
caffe_root            Please Input CAFFE HOME PATH(eg. /home/user/caffe)

optional arguments:
-h, --help            show this help message and exit
--gpus GPUS           Define GPU device ID
--run-soon            Directly Train Script Excuting
--network-model-name NETWORK_MODEL_NAME
                      Choose Applying Network Type(ex, lenet, resnet20,
                      resnet32 etc)

6. Test

* Using Script
cd $CAFFE_HOME/examples/traffic_sign_recognition
python test.py

usage: test.py [-h] [--gpus GPUS] [--run-soon]
             [--network-model-name NETWORK_MODEL_NAME]
             caffe_root

positional arguments:
caffe_root            Please Input CAFFE HOME PATH(eg. /home/user/caffe)

optional arguments:
-h, --help            show this help message and exit
--gpus GPUS           Define GPU device ID
--run-soon            Directly Train Script Excuting
--network-model-name NETWORK_MODEL_NAME
                      Choose Applying Network Type(ex, lenet, resnet20,
                      resnet32 etc)