YOLOV3

Introduction

This is my own YOLOV3 written in pytorch, and is also the first time i have reproduced a object detection model.The dataset used is PASCAL VOC. The eval tool is the voc2010. Now the mAP gains the goal score.

Subsequently, i will continue to update the code to make it more concise , and add the new and efficient tricks.

Note : Now this repository supports the model compression in the new branch model_compression

Results

Note :

• YOLOV3-448-544 means train image size is 448 and test image size is 544. "*" means the multi-scale.
• mAP(mine)'s format is (use_difficult mAP | no_difficult mAP).
• In the test, the nms threshold is 0.5(except the last one) and the conf_score is 0.01.others nms threshold is 0.45(0.45 will increase the mAP)
• Now only support the single gpu to train and test.

Environment

• Nvida GeForce RTX 2080 Ti
• CUDA10.0
• CUDNN7.0
• ubuntu 16.04
• python 3.5

# install packages
pip3 install -r requirements.txt --user

Brief

• Data Augment (RandomHorizontalFlip, RandomCrop, RandomAffine, Resize)
• Step lr Schedule
• Multi-scale Training (320 to 640)
• focal loss
• GIOU
• Label smooth
• Mixup
• cosine lr
• Multi-scale Test and Flip

Prepared work

1、Git clone YOLOV3 repository

git clone https://github.com/Peterisfar/YOLOV3.git

update the "PROJECT_PATH" in the params.py.

2、Download dataset

• Download Pascal VOC dataset : VOC 2012_trainval 、VOC 2007_trainval、VOC2007_test. put them in the dir, and update the "DATA_PATH" in the params.py.
• Convert data format : Convert the pascal voc *.xml format to custom format (Image_path0   xmin0,ymin0,xmax0,ymax0,class0   xmin1,ymin1...)

cd YOLOV3 && mkdir data
cd utils
python3 voc.py # get train_annotation.txt and test_annotation.txt in data/

3、Download weight file

• Darknet pre-trained weight : darknet53-448.weights
• This repository test weight : best.pt

Make dir weight/ in the YOLOV3 and put the weight file in.

Train

Run the following command to start training and see the details in the config/yolov3_config_voc.py

WEIGHT_PATH=weight/darknet53_448.weights

CUDA_VISIBLE_DEVICES=0 nohup python3 -u train.py --weight_path $WEIGHT_PATH --gpu_id 0 > nohup.log 2>&1 &

Notes:

• Training steps could run the "cat nohup.log" to print the log.
• It supports to resume training adding --resume, it will load last.pt automaticly.

Test

You should define your weight file path WEIGHT_FILE and test data's path DATA_TEST

WEIGHT_PATH=weight/best.pt
DATA_TEST=./data/test # your own images

CUDA_VISIBLE_DEVICES=0 python3 test.py --weight_path $WEIGHT_PATH --gpu_id 0 --visiual $DATA_TEST --eval

The images can be seen in the data/

TODO

• Mish
• OctConv
• Custom data

Reference

• tensorflow : https://github.com/Stinky-Tofu/Stronger-yolo
• pytorch : https://github.com/ultralytics/yolov3
• keras : https://github.com/qqwweee/keras-yolo3