Mmrotate rotation target detection framework usage record

mmrotate Rotating target detection framework usage record

Recently, there is a need for rotating target detection , In the process of finding rotating target detection algorithm , It is found that some people have evaluated various rotating target detection frameworks in advance , The result is mmrorate The best effect . Record this time mmrorate Experimental process .
The experimental records will be continuously updated ！
Main reference ：
1. Evaluation reference ：

https://blog.csdn.net/FL1623863129/article/details/123408354?utm_medium=distribute.pc_relevant.none-task-blog-2_defaultbaidujs_title~default-4-123408354-blog-124335060.pc_relevant_multi_platform_whitelistv1&spm=1001.2101.3001.4242.3&utm_relevant_index=7

2. Implementation process reference

https://blog.csdn.net/Sylvia_Lan/article/details/124008837
https://github.com/open-mmlab/mmrotate
https://sxj731533730.blog.csdn.net/article/details/124107625?spm=1001.2101.3001.6650.4&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7Edefault-4-124107625-blog-124008837.pc_relevant_default&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7Edefault-4-124107625-blog-124008837.pc_relevant_default&utm_relevant_index=6
https://blog.csdn.net/qq_43581224/article/details/123838415

One 、 Configuration environment

1. At present, the computer environment
（1）Ubuntu20.04
（2）Cuda==11.2+cudnn
2. Download the code to create a virtual environment
（1） Download code

https://github.com/open-mmlab/mmrotate

（2） Build a virtual environment
use pycharm Open the code folder , And add virtual environment

（3） install pytorch
Get into pytorch Official website , Find your own version , Download and install .
notes ： If there is a small partner who installs the same as me Cuda11.2, stay pytorch Official website , When the corresponding version cannot be found , But try installing cuda11.3 Corresponding pytorch==1.11.0 edition .
(4) install mmrorate

pip install openmim
mim install mmrotate

Two 、 test

1. Download weight
Weight model set

https://download.openmmlab.com/mmrotate/v0.1.0/kfiou/r3det_kfiou_ln_r50_fpn_1x_dota_oc/r3det_kfiou_ln_r50_fpn_1x_dota_oc-8e7f049d.pth

After downloading , stay mmrorate Under the folder , newly build model Folder , And put in model In the folder .

2. function demo
Run instruction format ：

python demo/image_demo.py \
    ${
    IMG_ROOT} \
    ${
    CONFIG_FILE} \
    ${
    CHECKPOINT_FILE}

IMG_ROOT： For the image to be detected
CONFIG_FILE： For profile
CHECKPOINT_FILE： For trained weights .

pycharm terminal cd Avenue demo Folder , Input

python3 image_demo.py demo.jpg ../configs/kfiou/r3det_kfiou_ln_r50_fpn_1x_dota_oc.py ../model/r3det_kfiou_ln_r50_fpn_1x_dota_oc-8e7f049d.pth

thus , It can be determined that the overall environment is ok .