A minimal PyTorch implementation of YOLOv3, with support for training, interface & evalution.
To train on custom dataset please visit my another [GitRepo].
Standard YOLO V3 Approch is modified, and the table shows what change has been made.
| Standard YOLO Approch | Self-Modified YOLO Approch |
|---|---|
| 1. Object Detection Process | 1. Object Detection Process |
| ... 1.1. Localization | ... 1.1. Localization |
| ... 1.2. Class Prediction | ... 1.2. Class Prediction |
| 2. Thresholding | 2. Thresholding |
| 3. Non max suppression with respect to Class | 3. Non max suppression irrespective of class label |
| 4. Bounding Box Labelling |
- Explanation : explanation.md
- You can check the full report of this project here... [FACE-MASK DETECTION USING YOLO V3 ARCHITECTURE.pdf]
> git clone https://github.com/NisargPethani/Face-Mask-Detection-using-YOLO-v3.git
> cd Face-Mask-Detection-using-YOLO-v3/
> pip install -r requirements.txt
Download checkpoint From [GoogleDrive]
Copy yolov3_ckpt_35.pth into checkpoints/
Evaluates the model.
> python validate.py --weights_path checkpoints/yolov3_ckpt_35.pth
Average Precisions:
| Class | AP |
|---|---|
| Class '0' (MASK) | 73.0 |
| Class '1' (NO_MASK) | 83.3 |
mAP: 78.19
> python cam_detect.py --weights_path checkpoints/yolov3_ckpt_35.pth
Some Screen-shots of Real-Time Detection is shown below
Move inmages to testing/input/images
> python image_detect.py --image_folder testing/input/images --weights_path checkpoints/yolov3_ckpt_35.pth
Make new directory with name: 'videos' in testing/input
Move videos to testing/input/videos
> python video_detect.py --image_folder testing/input/videos --weights_path checkpoints/yolov3_ckpt_35.pth
Following YouTube video shows the output.
[Paper] [Project Webpage] [Authors' Implementation]
@article{yolov3,
title={YOLOv3: An Incremental Improvement},
author={Redmon, Joseph and Farhadi, Ali},
journal = {arXiv},
year={2018}
}
Also Help is taken from [GitRepo]