Focal Loss for Dense Rotation Object Detection

Abstract

This repo is based on Focal Loss for Dense Object Detection, and it is completed by YangXue.

This is the baseline work of R³Det, paper link: R³Det: Refined Single-Stage Detector with Feature Refinement for Rotating Object.

Performance

More results and trained models are available in the MODEL_ZOO.md.

DOTA1.0

R³Det*: R³Det with two refinement stages
The performance of all models comes from the source paper.

Visualization

My Development Environment

docker images: docker pull yangxue2docker/yx-tf-det:tensorflow1.13.1-cuda10-gpu-py3
1、python3.5 (anaconda recommend)
2、cuda 10.0
3、opencv(cv2)
4、tfplot 0.2.0 (optional)
5、tensorflow 1.13

IoU-smooth L1 Loss

SCRDet: Towards More Robust Detection for Small, Cluttered and Rotated Objects (ICCV2019)

Download Model

Pretrain weights

1、Please download resnet50_v1, resnet101_v1 pre-trained models on Imagenet, put it to data/pretrained_weights.
2、(Recommend in this repo) Or you can choose to use a better backbone, refer to gluon2TF.

• Baidu Drive, password: 5ht9.

• Google Drive

Compile

cd $PATH_ROOT/libs/box_utils/cython_utils
python setup.py build_ext --inplace (or make)

cd $PATH_ROOT/libs/box_utils/
python setup.py build_ext --inplace

Train

1、If you want to train your own data, please note:

(1) Modify parameters (such as CLASS_NUM, DATASET_NAME, VERSION, etc.) in $PATH_ROOT/libs/configs/cfgs.py
(2) Add category information in $PATH_ROOT/libs/label_name_dict/lable_dict.py     
(3) Add data_name to $PATH_ROOT/data/io/read_tfrecord.py

2、Make tfrecord
For DOTA dataset:

cd $PATH_ROOTdataioDOTA
python data_crop.py

cd $PATH_ROOT/data/io/  
python convert_data_to_tfrecord.py --VOC_dir='/PATH/TO/DOTA/' 
                                   --xml_dir='labeltxt'
                                   --image_dir='images'
                                   --save_name='train' 
                                   --img_format='.png' 
                                   --dataset='DOTA'

3、Multi-gpu train

cd $PATH_ROOT/tools
python multi_gpu_train.py

Eval

cd $PATH_ROOT/tools
python test_dota.py --test_dir='/PATH/TO/IMAGES/'  
                    --gpus=0,1,2,3,4,5,6,7

Tensorboard

cd $PATH_ROOT/output/summary
tensorboard --logdir=.

Reference

1、https://github.com/endernewton/tf-faster-rcnn
2、https://github.com/zengarden/light_head_rcnn
3、https://github.com/tensorflow/models/tree/master/research/object_detection
4、https://github.com/fizyr/keras-retinanet