SOC-YOLO: Small and Overlapping Worker Detection at Construction Sites

This is the implementation of our Automation in Construction research paper:

Minsoo Park, Dai Quoc Tran, Jinyeong Bak, Seunghee Park - SOC-YOLO: Small and Overlapping Worker Detection at Construction Sites

Installation

This code is based on YOLOv5. Please install the code according to the YOLOv5 tutorial first.

SOC-YOLO Implementation

The updated code for implementing SOC-YOLO is displayed below:

1. Updated DIoU-NMS in general.py
2. Updated DIoU loss function in loss.py
3. P2:Feature-level expansion is added in P2.yaml
4. SoftPool use softpool.py and modify SPPF in common.py
5. We modify and add Weighted-tiplet attetion in common.py with variable hyper parameters alpha, beta, gamma, and advanced yaml is added in P2_Triple.yaml

Reference

• https://github.com/ultralytics/yolov5
• Zheng, Zhaohui, et al. "Distance-IoU loss: Faster and better learning for bounding box regression." Proceedings of the AAAI conference on artificial intelligence. Vol. 34. No. 07. 2020.
• Stergiou, Alexandros, Ronald Poppe, and Grigorios Kalliatakis. "Refining activation downsampling with SoftPool." Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021.
• Misra, Diganta, et al. "Rotate to attend: Convolutional triplet attention module." Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 2021.