YOLOv3_KITTI

Prepare dataset

img dataset

label dataset

train : val = 9 : 1 $\simeq$ 6881 : 600

mkdir eval train
mkdir eval/Annotations eval/ImageSets eval/JPEGImages
mv data_object_image_2/training/image_2/*{006881..007480}.png eval/JPEGImages/
mv kitti_training_annotations_yolo/*{006881..007480}.txt kitti_dataset/eval/Annotations/

mkdir train/Annotations train/ImageSets train/JPEGImages
mv data_object_image_2/training/image_2/*.png train/JPEGImages/
mv kitti_training_annotations_yolo/*.txt kitti_dataset/train/Annotations/

ImageSets

find ./train/Annotations/ -type f -name "*.txt" > ./train/ImageSets/train.txt

find ./eval/Annotations/ -type f -name "*.txt" > ./eval/ImageSets/eval.txt

Convert label format from KITTI to YOLO

KITTI format

type	클래스
truncated	이미지 밖으로 나간 정도
occluded	다른 객체에 가려진 정도
alpha	객체 각도
bbox	x_min y_min x_max y_max
dimensions
location
rotation_y
score

YOLO format

├── Annotations
├── ImageSets
└── JPEGImages

class c_x c_y w h

Preprocessing

transform

train

python main.py --mode train --cfg yolov3_kitti.cfg

forward

test

epoch_5 000007

epoch_7 000335

Error

pip install pyqt5-tools pyqt5
pip uninstall pyqt5-tools pyqt5 -y

Ref

KITTI dataset
Download left color images of object data set (12 GB)
Download training labels of object data set (5 MB)
convert2Yolo
imgaug