# Caffe SSD 目标检测

基于 Ubuntu16.04 和 Python2.7。

## 1.Installation

按《Ubuntu 初始配置》中的方法来安装开发工具。

1. Get the code. We will call the directory that you cloned Caffe into $CAFFE_ROOT

    ```bash
    git clone https://github.com/weiliu89/caffe.git
    cd caffe
    git checkout ssd
    ```

2. 修改 Makefile 文件：

    ```Makefile
    -LIBRARIES += glog gflags protobuf boost_system boost_filesystem boost_regex m hdf5_hl hdf5
    +LIBRARIES += glog gflags protobuf boost_system boost_filesystem boost_regex m hdf5_serial_hl hdf5_serial
    ```

3. Build the code. Please follow Caffe instruction to install all necessary packages and build it.

    ```bash
    # Modify Makefile.config according to your Caffe installation.
    cp Makefile.config.example Makefile.config
    # 通过去除 Makefile.config 中“CPU_ONLY := 1”前面的 “#” 号可选择编译 CPU 版本的 Caffe
    make -j8
    # Make sure to include $CAFFE_ROOT/python to your PYTHONPATH.
    make py
    make test -j8
    # (Optional)
    make runtest -j8
    cd python
    for req in $(cat requirements.txt); do pip install $req; done
    pip install -r requirements.txt
    export PYTHONPATH=$PYTHONPATH:$CAFFE_ROOT/python
    cd ..
    make pycaffe
    # 验证 pycaffe 接口
    python
    >>>import caffe
    >>>exit()
    ```

## 2.Preparation

1. Download [fully convolutional reduced (atrous) VGGNet](https://gist.github.com/weiliu89/2ed6e13bfd5b57cf81d6). By default, we assume the model is stored in $CAFFE_ROOT/models/VGGNet/
2. Download [VOC2007 and VOC2012 dataset](https://pjreddie.com/projects/pascal-voc-dataset-mirror/). By default, we assume the data is stored in $HOME/data/

    ```bash
    # Download the data.
    cd $HOME/data
    wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
    wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
    wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
    # Extract the data.
    tar -xvf VOCtrainval_11-May-2012.tar
    tar -xvf VOCtrainval_06-Nov-2007.tar
    tar -xvf VOCtest_06-Nov-2007.tar
    ```

3. 修改 data/VOC0712/create_list.sh 文件：

    ```bash
    -root_dir=$HOME/data/VOCdevkit/
    +root_dir=/mnt/workspace/caffe/VOC/VOCdevkit/
    ```

4. 修改 data/VOC0712/create_data.sh 文件：

    ```bash
    -data_root_dir="$HOME/data/VOCdevkit"
    +data_root_dir="/mnt/workspace/caffe/VOC/VOCdevkit"
    ```

5. Create the LMDB file.

    ```bash
    cd $CAFFE_ROOT
    # Create the trainval.txt, test.txt, and test_name_size.txt in data/VOC0712/
    ./data/VOC0712/create_list.sh
    # You can modify the parameters in create_data.sh if needed.
    # It will create lmdb files for trainval and test with encoded original image:
    #   - $HOME/data/VOCdevkit/VOC0712/lmdb/VOC0712_trainval_lmdb
    #   - $HOME/data/VOCdevkit/VOC0712/lmdb/VOC0712_test_lmdb
    # and make soft links at examples/VOC0712/
    ./data/VOC0712/create_data.sh
    ```

## 3.Train/Eval

1. 修改 examples/ssd/ssd_pascal.py 文件：

    ```python
    -if num_gpus > 0:
    -  batch_size_per_device = int(math.ceil(float(batch_size) / num_gpus))
    -  iter_size = int(math.ceil(float(accum_batch_size) / (batch_size_per_device * num_gpus)))
    -  solver_mode = P.Solver.GPU
    -  device_id = int(gpulist[0])
    +#if num_gpus > 0:
    +#  batch_size_per_device = int(math.ceil(float(batch_size) / num_gpus))
    +#  iter_size = int(math.ceil(float(accum_batch_size) / (batch_size_per_device * num_gpus)))
    +#  solver_mode = P.Solver.GPU
    +#  device_id = int(gpulist[0])
    ```

2. Train your model and evaluate the model on the fly.

    ```bash
    # It will create model definition files and save snapshot models in:
    #   - $CAFFE_ROOT/models/VGGNet/VOC0712/SSD_300x300/
    # and job file, log file, and the python script in:
    #   - $CAFFE_ROOT/jobs/VGGNet/VOC0712/SSD_300x300/
    # and save temporary evaluation results in:
    #   - $HOME/data/VOCdevkit/results/VOC2007/SSD_300x300/
    # It should reach 77.* mAP at 120k iterations.
    python examples/ssd/ssd_pascal.py
    ```

    If you don't have time to train your model, you can download a pre-trained model at [here](https://drive.google.com/open?id=0BzKzrI_SkD1_WVVTSmQxU0dVRzA).

3. Evaluate the most recent snapshot.

    ```bash
    # If you would like to test a model you trained, you can do:
    python examples/ssd/score_ssd_pascal.py
    ```

4. Test your model using a webcam. Note: press "esc" to stop.

    ```bash
    # If you would like to attach a webcam to a model you trained, you can do:
    python examples/ssd/ssd_pascal_webcam.py
    ```

    [Here](https://drive.google.com/file/d/0BzKzrI_SkD1_R09NcjM1eElLcWc/view) is a demo video of running a SSD500 model trained on [MSCOCO](http://mscoco.org/) dataset.

5. Check out [examples/ssd_detect.ipynb](https://hub.fastgit.org/weiliu89/caffe/blob/ssd/examples/ssd_detect.ipynb) or [examples/ssd/ssd_detect.cpp](https://hub.fastgit.org/weiliu89/caffe/blob/ssd/examples/ssd/ssd_detect.cpp) on how to detect objects using a SSD model. Check out [examples/ssd/plot_detections.py](https://hub.fastgit.org/weiliu89/caffe/blob/ssd/examples/ssd/plot_detections.py) on how to plot detection results output by ssd_detect.cpp.
6. To train on other dataset, please refer to data/OTHERDATASET for more details. We currently add support for COCO and ILSVRC2016. We recommend using [examples/ssd.ipynb](https://hub.fastgit.org/weiliu89/caffe/blob/ssd/examples/ssd_detect.ipynb) to check whether the new dataset is prepared correctly.

## 4.Models

We have provided the latest models that are trained from different datasets. To help reproduce the results in [Table 6](https://arxiv.org/pdf/1512.02325v4.pdf), most models contain a pretrained `.caffemodel` file, many `.prototxt` files, and python scripts.

1. PASCAL VOC models:
   * 07+12: [SSD300*](https://drive.google.com/open?id=0BzKzrI_SkD1_WVVTSmQxU0dVRzA), [SSD512*](https://drive.google.com/open?id=0BzKzrI_SkD1_ZDIxVHBEcUNBb2s)
   * 07++12: [SSD300*](https://drive.google.com/open?id=0BzKzrI_SkD1_WnR2T1BGVWlCZHM), [SSD512*](https://drive.google.com/open?id=0BzKzrI_SkD1_MjFjNTlnempHNWs)
   * COCO[1]: [SSD300*](https://drive.google.com/open?id=0BzKzrI_SkD1_NDlVeFJDc2tIU1k), [SSD512*](https://drive.google.com/open?id=0BzKzrI_SkD1_TW4wTC14aDdCTDQ)
   * 07+12+COCO: [SSD300*](https://drive.google.com/open?id=0BzKzrI_SkD1_UFpoU01yLS1SaG8), [SSD512*](https://drive.google.com/open?id=0BzKzrI_SkD1_X3ZXQUUtM0xNeEk)
   * 07++12+COCO: [SSD300*](https://drive.google.com/open?id=0BzKzrI_SkD1_TkFPTEQ1Z091SUE), [SSD512*](https://drive.google.com/open?id=0BzKzrI_SkD1_NVVNdWdYNEh1WTA)

2. COCO models:
   * trainval35k: [SSD300*](https://drive.google.com/open?id=0BzKzrI_SkD1_dUY1Ml9GRTFpUWc), [SSD512*](https://drive.google.com/open?id=0BzKzrI_SkD1_dlJpZHJzOXd3MTg)

3. ILSVRC models:
   * trainval1: [SSD300*](https://drive.google.com/open?id=0BzKzrI_SkD1_a2NKQ2d1d043VXM), [SSD500](https://drive.google.com/open?id=0BzKzrI_SkD1_X2ZCLVgwLTgzaTQ)

[1]We use [`examples/convert_model.ipynb`](https://github.com/weiliu89/caffe/blob/ssd/examples/convert_model.ipynb) to extract a VOC model from a pretrained COCO model.