/ Machine Learning

Focal Loss for Dense Rotation Object Detection

Focal Loss for Dense Rotation Object Detection

RetinaNet_Tensorflow_Rotation

Focal Loss for Dense Rotation Object Detection

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

Performance

Model Backbone Training data Val data mAP GPU Image/GPU Anchor Reg. Loss configs
RetinaNet (baseline) ResNet50_v1 600->800 DOTA1.0 trainval DOTA1.0 test 62.79 8X GeForce RTX 2080 Ti 1 H smooth L1 cfgs_res50_dota_v11.py
RetinaNet (baseline) ResNet50_v1 600->800 DOTA1.0 trainval DOTA1.0 test 62.02 1X GeForce RTX 2080 Ti 1 R smooth L1 cfgs_res50_dota_v1.py
RetinaNet (baseline) ResNet50_v1 600->800 DOTA1.0 trainval DOTA1.0 test 63.45 1X GeForce RTX 2080 Ti 1 R IoU-smooth L1 cfgs_res50_dota_v8.py

demo2

My Development Environment

1、python3.5 (anaconda recommend)
2、cuda9.0
3、opencv(cv2)
4、tfplot (optional)
5、tensorflow >= 1.12

IoU-smooth L1 Loss

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

example

iou_smooth_l1_loss

Download Model

Pretrain weights

1、Please download resnet50_v1, resnet101_v1 pre-trained models on Imagenet, put it to data/pretrained_weights.
2、Or you can choose to use a better backbone, refer to gluon2TF. Pretrain Model Link, password: 5ht9. (Recommend)

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_ROOT\data\io\DOTA
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=.

images

scalars

GitHub