Object Detection using MMDetection on KITTI Dataset

Introduction

Welcome to this repository, which focuses on:

• Training a 2D object detector for the KITTI dataset.
• Utilizing the powerful "MMDetection" framework for 2D object detection.

For more detailed information, please refer to the official MMDetection documentation: https://mmdetection.readthedocs.io/en/latest/

1. Getting Started

Clone the repo:

git clone https://github.com/mingkyun/kitti_mmdetection

Download Kitti Dateset from following link
https://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=2d

Download left color images of object data set (12 GB)
Download training labels of object data set (5 MB)

Then put both both files at ./KITTI_MMDETECTION/data/

└── KITTI_MMDETECTION
    └── data
        ├── data_object_image_2.zip
        └── data_object_label_2.zip

2. Requirements

python>=3.7
PyTorch>=1.8
CUDA >=9.2
mmengine>=0.7.0
mmcv >= 2.0.0

Install all dependent libraries:

install/install_packages.sh

or

python install/install_mmdetection.py

for users who do not have administrator privileges

3. Preprocessing

Execute following command for all preprocessing procedure at once

preprocessing/preprocess.sh

or execute following commands at /kitti_mmdetection/

(1) unzip files

python preprocessing/unzip.py   

(2) split to train/val sets (default split rate 0.8)

python preprocessing/split.py --split 0.9 

(3) transfer kitti annotation to COCO format

python preprocessing/toCOCO.py

After all process done, your directory should look like this

KITTI_MMDETECTION
├── data
│   ├── data_object_image_2.zip
│   ├── data_object_label_2.zip
│   ├── image
│   │   ├── image_file_1.png
│   │   ├── image_file_2.png
│   │   └── ...
│   ├── label
│   │   ├── label_file_1.txt
│   │   ├── label_file_2.txt
│   │   └── ...
│   ├── train
│   │   ├── image
│   │   │   ├── image_file_1.png
│   │   │   ├── image_file_2.png
│   │   │   └── ...
│   │   ├── label
│   │   │   ├── label_file_1.txt
│   │   │   ├── label_file_2.txt
│   │   │   └── ...
│   │   └── coco
│   │       └── kitti_coco_format_train.json
│   └── val
│       ├── image
│       │   ├── image_file_3.png
│       │   ├── image_file_8.png
│       │   └── ...
│       ├── label
│       │   ├── label_file_3.txt
│       │   ├── label_file_8.txt
│       │   └── ...
│       └── coco
│           └── kitti_coco_format_val.json
├── mmdetection
├── install
├── preprocessing
└── train

4. Train

(Recommended) Follow procedure at train/build_config.ipynb

After making config file, train model by following command (Modify your path)

python mmdetection/tools/train.py {config file path} [--options]

Example

python mmdetection/tools/train.py mmdetection/configs/efficientnet/retinanet_effb3_fpn_8xb4-crop896-1x_kitti.py

5. Test

After making config file, train model by following command (Modify your path)

python mmdetection/tools/test.py {config file path}{checkpoint file path} [--options]

You can also visualize your model's output at train/test.ipynb

For those who want to use tracking dataset as test set, you can follow guidelines in the build_config file

Citation

@article{mmdetection,
  title   = {{MMDetection}: Open MMLab Detection Toolbox and Benchmark},
  author  = {Chen, Kai and Wang, Jiaqi and Pang, Jiangmiao and Cao, Yuhang and
             Xiong, Yu and Li, Xiaoxiao and Sun, Shuyang and Feng, Wansen and
             Liu, Ziwei and Xu, Jiarui and Zhang, Zheng and Cheng, Dazhi and
             Zhu, Chenchen and Cheng, Tianheng and Zhao, Qijie and Li, Buyu and
             Lu, Xin and Zhu, Rui and Wu, Yue and Dai, Jifeng and Wang, Jingdong
             and Shi, Jianping and Ouyang, Wanli and Loy, Chen Change and Lin, Dahua},
  journal= {arXiv preprint arXiv:1906.07155},
  year={2019}
}

License

This project is released under the Apache 2.0 license.