Note: Checkout the master branch. As of now, all the code is committed in that branch and is not merged to main.
For detailed approach, please visit here: Google doc
-
I tried to make bbox using only opencv. But it was very difficult to approximate the contours and make a convex hull around each item.
-
I used pre-trained model YOLO version 8 but that model was not trained for custom object detection. So that didn't work.
-
Augmented the sample images using labelImg library as there were very less number of images to train the YOLO model.
-
Trained the YOLO model on augmented images, and made inference:
b. For the remaining results do check the 'results custom yolov8' and 'supporting screenshots' directory
- Clone this repository or download it.
- make a virtual environment in python 3.7
- install the all libraries reqirements.txt
- Run yolo_custom.py file.
- You will see the 2 window as outcome: one will be input image.. another one will be output image..