obeychoi0120/YOLOv5-GradCAM

SKKU S-HERO Capstone Projects with PCB detection using Yolov5&Grad-CAM

PythonMIT

YOLOv5-GradCAM

SKKU S-HERO Capstone Project with PCB defects detection
using YOLOv5 & GradCAM (Pytorch)

YOLOv5 Source: https://github.com/ultralytics/yolov5
Grad-CAM Source: https://github.com/jacobgil/pytorch-grad-cam

Usage

1. Object Detection using YOLOv5

Check 'yolov5' folder, read usage first
Structure your files as yolov5's input
Run train.py for training
and detect.py to save detection results (if you're not interesting at interpreting results, you can skip from here.)

2. Interpret classification results using grad-CAM

2-1. Train YOLOv5_classifier

We're interpreting classification results only, not including detection
Run train.py for training yolo_classifier: submodel of yolov5 with detecting architecture removed
Make sure you use the same structure as yolov5 (s, m, l, x).
Classification results will be SAME between original yolo and yolo_classifier

2-2. Run cam.py

In 'pytorch-grad-cam' folder
Modify 'model' to the same model as you used before
Modify 'ckpt' to your own trained weight
Run script(check pytorch-grad-cam usage)

Detection & GradCAM results