2019 第五届“四维图新”杯创新大赛自动驾驶视觉综合感知算法赛 detection baseline

testA object detection map: 0.22

testA semantic segmentation miou: 0.46

author： zhengye

for the semantic segmentation part, please refer to my teammate's git repo: https://github.com/SHERLOCKLS/datafountain_siweituxin_autodriver_seg

环境配置

请按照mmdetection说明进行安装配置

数据准备

将训练数据所有训练图片放置于 data/siweituxin/train_image

两批次的数据label分别位于'data/dataset1/train.txt' 和 'data/DF_1018/train.txt'

合并label： python tools/convert_datasets/merage_txt_label.py

txt转coco json： python tools/convert_datasets/trans_txt2json.py
模型训练

请依据mmdetection说明依据自身显卡情况线性调整lr，这里以4卡为例

CUDA_VISIBLE_DEVICES=0,1,2,3 ./tools/dist_train.sh config/siweituxin/faster_rcnn_r50_fpn.py 4

数据准备

将测试数据所有训练图片放置于 data/siweituxin/test_images
inference

CUDA_VISIBLE_DEVICES=0 python tools/infer_siweituxin.py config/siweituxin/faster_rcnn_r50_fpn.py /work_dirs/faster_rcnn_r50_fpn/latest.pth --out det.txt

This repo is currently maintained by Ye Zheng (@zhengye1995).

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
configs		configs
mmdet		mmdet
tests		tests
tools		tools
LICENSE		LICENSE
README.md		README.md
pytest.ini		pytest.ini
requirements.txt		requirements.txt
setup.py		setup.py