This is the object detection pipeline of AUV IITK, trained on YOLOv3 a real-time object detection DL model.
The custom dataset was made by annotating the images and storing the labels in YOLO format using OpenLabeling tool.
Detailed training process can be found in YOLOv3training notebook.
The last saved weights can be found in the here.
Current accuracy of the model trained on two classes gate and buoy is 85.7% which can be further increased by increasing the number of iterations.
The training results for 700 and 1000 iterations can be found in the Results folder.