• simpledet 的配置

• 1. 通过 docker 配置 simpledet

• 1.1 系统要求

  ubuntu16.04

  python >=3.5

• 1.2 下载 docker 镜像

  匹配的版本为 ubuntu16.04， cuda9.0， cudnn7， python3。

  https://gitlab.com/nvidia/cuda/blob/ubuntu16.04/9.0/devel/cudnn7/Dockerfile

• 1.3 运行 docker

  nvidia-docker run -v $HOST-SIMPLEDET-DIR:$CONTAINER-WORKDIR -it nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04 bash

• 1.4 安装所需环境

# Install dependency
sudo apt-get update
sudo apt-get install -y build-essential git
sudo apt-get install -y libopenblas-dev

• 1.5 下载 simpledet 和 pycocotools, mxnext 项目

git clone <https://github.com/TuSimple/simpledet.git>
cd /path/to/simpledet
make

# Install a patched cocotools for python3
git clone <https://github.com/RogerChern/cocoapi>
cd cocoapi/PythonAPI
python3 setup.py install

# setup mxnext, a wrapper of mxnet symbolic API
cd /path/to/simpledet
git clone <https://github.com/RogerChern/mxnext>

• 1.6 安装mxnet

# Specify simpledet directory
export SIMPLEDET_DIR=/path/to/simpledet
export COCOAPI_DIR=/path/to/cocoapi

git clone https://github.com/apache/incubator-mxnet mxnet
cd mxnet
git checkout 1.3.1
git submodule init
git submodule update
echo "USE_OPENCV = 0" >> ./config.mk
echo "USE_BLAS = openblas" >> ./config.mk
echo "USE_CUDA = 1" >> ./config.mk
echo "USE_CUDA_PATH = /usr/local/cuda" >> ./config.mk
echo "USE_CUDNN = 1" >> ./config.mk
echo "USE_NCCL = 1" >> ./config.mk
echo "USE_DIST_KVSTORE = 1" >> ./config.mk
cp -r $SIMPLEDET_DIR/operator_cxx/* src/operator/
mkdir -p src/coco_api
cp -r $COCOAPI_DIR/common src/coco_api/
make -j
cd python
python3 setup.py install

• 2. 使用 coco 测试集进行模型测试

• 2.1 下载模型

  在 simpledet 目录下新建 experiments 目录，并将下载好的模型（https://github.com/TuSimple/simpledet/tree/master/models/tridentnet）放至该路径下，如

experiments/
    tridentnet_r101v2c4_c5_1x/
        checkpoint-0006.params
        checkpoint-symbol.json
        log.txt
        coco_minival2014_result.json

• 2.2 构建 coco roidb 测试集，将coco数据集按以下目录结构进行存放

data/
    coco/
        annotations/
            instances_train2014.json
            instances_valminusminival2014.json
            instances_minival2014.json
            image_info_testdev2017.json
        images/
            train2014
            val2014
            test2017

• 2.3 执行转换命令，例如：

python3 utils/generate_roidb.py --dataset coco --dataset-split train2014
python3 utils/generate_roidb.py --dataset coco --dataset-split valminusminival2014
python3 utils/generate_roidb.py --dataset coco --dataset-split minival2014
python3 utils/generate_roidb.py --dataset coco --dataset-split test-dev2017

• 2.4 测试

python3 detection_test.py --config config/detection_config.py

• 3. 单张图像的检测

详见 https://github.com/danpe1327/simpledet/blob/master/detect_image.py