AVOD

一，安装依赖

avod需要使用python3，所以要使用pip3安装依赖库

<span style="color:#000000">pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple sklearn
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple matplotlib
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple pandas
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple pillow
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple scipy
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple protobuf==3.2.0
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple numpy
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple scikit-build
pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple opencv-python
二，安装tensorflow-gpu
使用阿里镜像安装gpu版的tensorflow，注意版本要是1.13.0，tensorflow依赖protobuf和numpy,会自动再安装一遍,命令如下：
pip3 install --ignore-installed --upgrade tensorflow-gpu<code>==1.13.0</code> -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
pip忽略依赖命令：</span>

pip3 install package --no-dependencies

<span style="color:#000000">二，下载数据集
1，在/home/username/目录下创建目录/Kitti/object/
2，下载地址
<span style="color:#000080"><u><a data-cke-saved-href="https://link.zhihu.com/?target=http%3A//www.cvlibs.net/datasets/kitti/eval_object.php%3Fobj_benchmark%3D3d" href="https://link.zhihu.com/?target=http%3A//www.cvlibs.net/datasets/kitti/eval_object.php%3Fobj_benchmark%3D3d">https://link.zhihu.com/?target=http%3A//www.cvlibs.net/datasets/kitti/eval_object.php%3Fobj_benchmark%3D3d</a></u></span>
3，下载如下红框的文件

4，解压放到目录/Kitti/object/下

三，将avod目录加到环境变量中
export PYTHONPATH=$PYTHONPATH:'/home/vking/codespace/avod-master/wavedata'
export PYTHONPATH=$PYTHONPATH:'/home/vking/codespace/avod-master'

<code><span style="color:#121212">四，编译</span></code><code>.proto</code><code>文件</code>
<code><span style="color:#121212">protoc avod/protos/*.proto –python_out=.</span></code>
<code><span style="color:#121212">如果</span></code><code><span style="color:#121212">protoc</span></code><code><span style="color:#121212">版本不对，从如下地址下载对应版本</span></code>
<code><a data-cke-saved-href="https://github.com/protocolbuffers/protobuf/releases/tag/v3.2.0" href="https://github.com/protocolbuffers/protobuf/releases/tag/v3.2.0"><span style="color:#121212">https://github.com/protocolbuffers/protobuf/releases/tag/v3.2.0</span></a></code>
<code><span style="color:#121212">下载后解压，</span></code><code><span style="color:#121212">bin</span></code><code><span style="color:#121212">目录下有个</span></code><code><span style="color:#121212">protoc</span></code><code><span style="color:#121212">文件，使用该文件编译</span></code><code><span style="color:#121212">.proto</span></code><code><span style="color:#121212">文件</span></code>
<code><span style="color:#121212">/home/username/opensource/protoc-3.2.0/bin/protoc avod/protos/*.proto –python_out=.</span></code>
<code><span style="color:#121212">avod</span></code><code><span style="color:#121212">的</span></code><code><span style="color:#121212">proto</span></code><code><span style="color:#121212">文件中可能头部需要加</span></code><code><span style="color:#121212">syntax = "proto2";</span></code>

五，生成RPN所需数据
<code><span style="color:#121212">python3 scripts/preprocessing/gen_mini_batches.py</span></code>



<code><span style="color:#121212">六，训练</span></code>
<code>python3 avod/experiments/run_training.py --pipeline_config=avod/configs/pyramid_cars_with_aug_example.config  --device='0' –data_split='train'</code>
</span>

1，show_predictions_2d.py

 

KittiDatasetConfig对应文件avod/protos/kitti_dataset.proto， kitti_dataset.proto中default表示属性的默认值，dataset_dir表示数据集存储的目录，默认为~/Kitti/object，在KittiDataset类的构造函数中会使用 expanduser来对符号~进行转换

self.dataset_dir = os.path.expanduser(self.config.dataset_dir)

最终dataset_dir会变成/home/username/Kitti/object