USOA

Unified model of Semantic segmentation and Object detection for Autonomous driving

Paper Link : https://lib.dongguk.edu/search/media/url/CAT000001257272

This description is based on Win10 + Anaconda3

Enviroments

OS : Win10

Manage Syatem : Anaconda3

Package Version

• Python = 3.8.12
• Pytorch = 1.10.1
• Albumentation = 1.0.3
• OpenCV = 4.0.1
• NVIDIA Driver : 472.47

Train Dataset : BDD100K train set + Argumentation(My Paper)

Test Dataset : BDD100K Validation set

Image Size : 640 x 384

Object Detection Class : Vehicle(Car + Bus + Truck + Train )

Semantic Segmentation Class : Background, Road(Alternative + Direction)

Model Architecture

Inference Video (BDD100K Test)

Daytime : https://www.youtube.com/watch?v=rAvok4emD-8

Night : https://www.youtube.com/watch?v=m36-rhSQ4cI

Performance

Model	Size	AP(IOU=0.5)	mIoU	FPS(RTX3090)
YOLOP	640x384	76.5	91.5	46.51
USOA	640x384	76.79	92.57	41.49

Paper Link : (https://lib.dongguk.edu/search/media/url/CAT000001257272)

Inference Image Example

How to Install package

0. Update Your NVIDIA Graphic Driver 
1. Open Anaconda3 prompt
2. conda create -n usoa python=3.8.12
3. conda activate usoa
4. conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge
5. conda install -c anaconda cudnn
6. conda install -c conda-forge albumentations

Quick Start

How to doing?

0. git clone https://github.com/tmvhs1004/USOA or Download zip, Unzip, Change folder name USOA-main to USOA
1. Download trained-weight file (File Link : https://drive.google.com/file/d/1oZSQRVQztqOmNqRiVHUy4ZPkl5XkDfLi/view?usp=sharing )
2. Move trained-weight file to './USOA/Weight/END/ ' Folder 
3. Open Anaconda Prompt
4. Cd to USOA folder 
5. Enter the command 'python test.py'
6. Waiting for testing time
7. Inference image is saved in './USOA/Result/output/' folder

Training

How to doing?

0. git clone https://github.com/tmvhs1004/USOA or Download zip, Unzip, Change folder name USOA-main to USOA
1. Download Dataset (File Link) https://drive.google.com/file/d/1K26G7jKbrsHHoiZ6c-7QRUgo7wFfl5M6/view?usp=sharing
2. Unzip to './USOA/Data/' Folder 
3. Change code line 115 at Train.py file
  train_set = './Data/Example/' -> train_set = './Data/Train_640x384_refo+w2h2/'
4. Change hyper-parameter such as 'batch size' in Config.py file
5. Open Anaconda Prompt
6. Cd to USOA folder 
7. Enter the command 'python train.py'
8. Waiting for training time
9. The weight file is saved in './USOA/Weight/' folder

Testing

How to doing?

0. git clone https://github.com/tmvhs1004/USOA or Download zip, Unzip, Change folder name USOA-main to USOA
1. Download Dataset (File Link : https://drive.google.com/file/d/1Zhe58ERgCIkw9yzQRTR9fmGg7U0EzNuu/view?usp=sharing )
2. Unzip to './USOA/Data/' Folder 
3. Change code line 145 at Test.py file
  test_set = './Data/Example/' -> test_set = './Data/Test_640x384_refo/'
4. Download trained-weight file (File Link : https://drive.google.com/file/d/1oZSQRVQztqOmNqRiVHUy4ZPkl5XkDfLi/view?usp=sharing )
5. Move trained-weight file to './USOA/Weight/END/ ' Folder 
6. Change hyper-parameter such as 'batch size' in Config.py file
7. Open Anaconda Prompt
8. Cd to USOA folder 
9. Enter the command 'python test.py'
10. Waiting for testing time
11. Inference image is saved in './USOA/Result/output/' folder