TYPE WHATEVER FUKCING AWESOME TITILE YOU WANT

## OAN Reshape
Step by steps explain the crucial code

e.g. the oan.py under config

---

## OAN Reshape

Step by steps explain the crucial code

e.g. the oan.py under config

# 1 Installation

---

## 1.1 Install miniconda3 && pytorch

1.1.a create oan environment
```conda
conda create -n oan -y python=3.7
conda activate oan
```

1.1.b install pytorch
```conda
conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.7 -c pytorch -c nvidia
```

---

## 1.2 Verify the torch installation

1.2.a
Enter the python envrionment
```conda
(oan) PS C:\Users\Git\OAN> python
```

Output shows the python version is 3.7
```python
Python 3.7.16 (default, Jan 17 2023, 16:06:28) [MSC v.1916 64 bit (AMD64)] :: Anaconda, Inc. on win32
Type "help", "copyright", "credits" or "license" for more information.
```

---

## 1.2 Verify the torch installation

1.2.b
Under python prompt and check if the torch cuda is available
```python
>>> import torch
>>> torch.cuda.is_available()
True
>>> ^Z
```

---

## 1.3 BboxToolKit

Install BboxToolkit
```conda
cd BboxToolkit
pip install -v -e .  # or "python setup.py develop"
```

Output
```conda
  Running setup.py develop for BboxToolkit
    Running command python setup.py develop
    running develop
    running build_ext
    Creating c:\users\git\miniconda3\envs\oan\lib\site-packages\BboxToolkit.egg-link (link to .)
    Adding BboxToolkit 1.1 to easy-install.pth file

Installed c:\users\git\oan\bboxtoolkit
    C:\Users\Git\miniconda3\envs\oan\lib\site-packages\setuptools\command\easy_install.py:147: EasyInstallDeprecationWarning: easy_install command is deprecated. Use build and pip and other standards-based tools.      
      EasyInstallDeprecationWarning,
    C:\Users\Git\miniconda3\envs\oan\lib\site-packages\setuptools\command\install.py:37: SetuptoolsDeprecationWarning: setup.py install is deprecated. Use build and pip and other standards-based tools.
      setuptools.SetuptoolsDeprecationWarning,
Successfully installed BboxToolkit-1.1 cycler-0.11.0 fonttools-4.38.0 kiwisolver-1.4.5 matplotlib-3.5.3 packaging-23.2 pyparsing-3.1.1 python-dateutil-2.8.2
```

---

## 1.4 Install mmdetection
### a. Install mmcv
```python
pip install mmcv==0.6.2
```

Output
```conda
Building wheels for collected packages: mmcv
  Building wheel for mmcv (setup.py) ... done
  Created wheel for mmcv: filename=mmcv-0.6.2-cp37-cp37m-win_amd64.whl size=206407 sha256=be7c9c11eeac6776e58c8e20adab8d1b84ef9e2ad1a09cdf567be4fa532ea506
  Stored in directory: c:\users\git\appdata\local\pip\cache\wheels\00\00\f1\ea823230a52a2d9b12582aa59e10445f50a943b08bce9051d4
Successfully built mmcv
Installing collected packages: addict, zipp, typing-extensions, tomli, pyyaml, platformdirs, importlib-metadata, yapf, mmcv
  Attempting uninstall: typing-extensions
    Found existing installation: typing_extensions 4.3.0
    Uninstalling typing_extensions-4.3.0:
      Successfully uninstalled typing_extensions-4.3.0
Successfully installed addict-2.4.0 importlib-metadata-6.7.0 mmcv-0.6.2 platformdirs-3.11.0 pyyaml-6.0.1 tomli-2.0.1 typing-extensions-4.7.1 yapf-0.40.2 zipp-3.15.0
```

---

## 1.4 Install mmdetection
### b. Install build requirements and then install mmdetection. 
(We install our forked version of pycocotools via the github repo instead of pypi for better compatibility with our repo.)

```conda
# back to mmdetection dir
pip install -r requirements/build.txt
pip install mmpycocotools
pip install pillow==6.2.2
pip install -v -e .  # or "python setup.py develop"
```

---

## 1.5 Download Pytorch Model

|model|oan|dataset|ss|BaiDu disk|Google Drive|
|:-------:|:--:|:----:|:--:|:--:|:--:|
|faster rcnn oan|√|dota-v1.0|√|[key:fc4n](https://pan.baidu.com/s/1pEIfJo1DdKLQzCQEH6v1qQ?pwd=fc4n)|[model](https://drive.google.com/file/d/1Yl29PZ8yBCib1aLQiBAb6NYTyqRzrzY0/view?usp=share_link)|
|retinanet oan|√|dota-v1.0|√|[key:uq8b](https://pan.baidu.com/s/1GB9ztvoMgGe404wX-Mr4Kw?pwd=uq8b)|[model](https://drive.google.com/file/d/1Ltb_fyCycgncBhuVcAOsUjd0mlDgvcgZ/view?usp=sharing)|
|roi trans oan|√|dota-v1.0|√|[key:jyww](https://pan.baidu.com/s/1D2J26xssLAYD6xwUd_XlSg?pwd=jyww)|[model](https://drive.google.com/file/d/1YSy15XmEO1KM6N9h7eKoYOZ1av5G3Vo0/view?usp=share_link)|
|oriented rcnn r50 oan|√|dota-v1.0|√|[key:bpyb](https://pan.baidu.com/s/1_EfOUw0rmJvEePQia1nCcw?pwd=bpyb)|[model](https://drive.google.com/file/d/1WSjEhHezWy83AAiVgQeMItc6UrMBk3ps/view?usp=share_link)|
|oriented rcnn x50 oan|√|dota-v1.0|√|[key:umtg](https://pan.baidu.com/s/15W4oeKLJy5aLUnK2Sajn8A?pwd=umtg)|[model](https://drive.google.com/file/d/1ad_ZiYCufP7hbkAnntr99UUkm9dVWVGR/view?usp=share_link)|

Download the Pytorch model and save in OAN directory

```
OAN
├── PTH
│   ├──faster_rcnn_obb_r50_fpn_1x_dota10_oan.pth
│   ├──retinanet_obb_r50_fpn_1x_dota10_oan.pth
│   ├──
```

---

## 2 Run OAN
## 2.1 BboxToolKit spilt images

### Splitting images (for DOTA)
The DOTA images are too big to train. We need to split the image before training.

```shell
cd BboxToolkit/tools
tree
```

Output
```shell
└─split_configs
    ├─dota1_0
    ├─dota1_5
    └─dota2_0
```

---

The Dota split config json files. (E.g. ms_test.json)
```json
{
  "nproc": 10,
  "load_type": "dota",
  "img_dirs": [
      "data/DOTA1_0/test/images/"
  ],
  "ann_dirs": null,
  "classes": null,
  "prior_annfile": null,
  "merge_type": "addition",
  "sizes": [
      1024
  ],
  "gaps": [
      200
  ],
  "rates": [
      1.0
  ],
  "img_rate_thr": 0.6,
  "iof_thr": 0.7,
  "no_padding": false,
  "padding_value": [
      104, 116, 124
  ],
  "save_dir": "data/split_ss_dota1_0/test/",
  "save_ext": ".png"
}
```

---

We shall put the dota dataset images under `BboxToolKit/tools/data`

And the output data will be saved in the `data/split_ss_dota1_0/test/` directory.

We can modify the path to the expected path of the The Dota split config json files. (E.g. ms_test.json)
```json
{
  "img_dirs": [
      "data/DOTA1_0/test/images/"
  ],
  "save_dir": "data/split_ss_dota1_0/test/",
  "save_ext": ".png"
}
```

---

Also, the sizes and gaps of the images can be changed in json config file.

```json
"sizes": [
1024
],
"gaps": [
200
],
```

we can refer to BboxToolKit github for more information

---

## 3 inference

```shell
# inference on a single image
python demo/huge_image_demo.py demo/large_image.jpg configs/oan/faster_rcnn_roitrans_r50_fpn_1x_dota10_oan.py P_PTH/faster_rcnn_roitrans_r50_fpn_1x_dota10_oan.pth  BboxToolkit/tools/split_configs/dota1_5/ss_test.json output.png
```

We can also select the specific image of DOTA datasets to inference

```shell
# inference on a single image
python demo/huge_image_demo.py /demo/P2805.png configs/oan/faster_rcnn_roitrans_r50_fpn_1x_dota10_oan.py P_PTH/faster_rcnn_roitrans_r50_fpn_1x_dota10_oan.pth  BboxToolkit/tools/split_configs/dota1_5/ss_test.json output.png
```

```shell
# inference on a single image
python demo/image_demo.py data/split_ss_dota1_0/test/images/P1971_0005.png configs/oan/faster_rcnn_obb_r50_fpn_1x_dota10_oan.py P_PTH/faster_rcnn_obb_r50_fpn_1x_dota10_oan.pth O_OUTPUT/OUT_P1971_0005.png
```

---

## 4 Testing

```shell
python tools/test.py configs/oan/faster_rcnn_orpn_r50_fpn_1x_dota10_ss_oan.py P_PTH/faster_rcnn_orpn_r50_fpn_1x_dota10_ss_oan.pth --format-only
```

---

the outputs of the testing shall be like this

```shell
dota_submission_dir$ ls
Task1_baseball-diamond.txt    Task1_harbor.txt         Task1_roundabout.txt         Task1_storage-tank.txt
Task1_basketball-court.txt    Task1_helicopter.txt     Task1_ship.txt               Task1_swimming-pool.txt
Task1_bridge.txt              Task1_large-vehicle.txt  Task1_small-vehicle.txt      Task1_tennis-court.txt
Task1_ground-track-field.txt  Task1_plane.txt          Task1_soccer-ball-field.txt  zip
```

---

e.g. ask1_ship.txt
```shell
P0298 0.9981546998023987 298.63 631.09 288.81 676.31 6.90 615.07 16.72 569.86
P0298 0.9966198205947876 810.12 112.31 806.46 131.43 749.94 120.60 753.61 101.48
P0298 0.9939270615577698 876.93 162.96 873.93 181.51 815.12 172.01 818.12 153.45
P0298 0.9933449029922485 880.58 128.89 876.50 147.04 824.11 135.27 828.19 117.12
P0298 0.993104875087738 721.14 130.94 716.89 149.22 663.58 136.83 667.83 118.55
P0298 0.9930562376976013 500.18 790.48 487.60 851.96 60.43 764.56 73.01 703.08
P0298 0.9925026893615723 820.13 341.33 815.88 359.16 763.03 346.55 767.28 328.72
P0298 0.9919962286949158 832.38 306.50 828.92 324.70 763.71 312.33 767.16 294.13
P0298 0.9904623627662659 702.75 552.19 688.91 614.42 232.03 512.78 245.88 450.54
```

---

## 5 Training

```shell
python tools/train.py configs/oan/faster_rcnn_obb_r50_fpn_1x_dota10_oan.py
```

.center[<img src="re-oan/oan_train.png" width="800">]

---

## 5 Training

The trained outputs will be like this

```shell
20230529_202330.log       epoch_11.pth  epoch_2.pth  epoch_5.pth  epoch_8.pth                               latest.pth
20230529_202330.log.json  epoch_12.pth  epoch_3.pth  epoch_6.pth  epoch_9.pth
epoch_10.pth              epoch_1.pth   epoch_4.pth  epoch_7.pth  faster_rcnn_obb_r50_fpn_1x_dota10_oan.py
```