Object-Detection-In-Low-light-Illumination

Capstone Project of Computer Vision Course

Members

Nguyen Hoang Dang - 20194423

Do Quoc An - 20194414

Ha Vu Thanh Dat - 20194424

Le Hai Son - 20194449

Enviroment

python 3.9
pytorch 1.10.0
mmcv 1.5.3
opencv-python tqdm

Pre-process

Step-1: Cd in "your_project_path", and do set-up process (see mmdetection if you want find details):

git clone https://github.com/dang-nh/Object-Detection-In-Night-Vision.git

cd "your project path"

pip install -r requirements.txt

Step-2: Download checkpoint and config file from goole drive

Step-3: Put checkpoint and config file to "checkpoints" folder and "config" folder like below folder structure

├── configs
│   └── yolov3_only_image_process.py
├── checkpoints
│   └── best_mAP_epoch_43.pth
├── infer.py # inference code
├── train.py # training code
├── test.py # testing code

Experiment Result

class	gts	dets	recall	ap
Bicycle	212	737	0.910	0.825
Boat	289	845	0.879	0.790
Bottle	282	1029	0.876	0.763
Bus	135	287	0.978	0.937
Car	597	1537	0.931	0.853
Cat	183	518	0.874	0.723
Chair	466	1886	0.848	0.717
Cup	366	966	0.869	0.794
Dog	207	637	0.928	0.813
Motorbike	233	721	0.854	0.771
People	1562	3860	0.908	0.825
Table	333	1491	0.772	0.575
mAP				0.782

Inference

Infer model with a set of image

python3 infer.py --input_dir [input directory path] --output_dir [output directory path] --checkpoint checkpoints/best_mAP_epoch_43.pth --cfg configs/yolov3_only_image_process.py

The code is largely borrow from mmdetection and unprocess, Thanks to their wonderful works