S23-AML-Assignment-1

Detection of street lights and security cameras

1. Dataset collection

The dataset was collected over two days with different lighting and weather.

2. Dataset annotation

The dataset was annotated using Roboflow, and can be accessed via the link.

3. Faster RCNN training

The faster RCNN model was trained using detectron2.

4. YOLOv8 training

The YOLO v8 nano was trained using ultralytics.

5. Faster RCNN and YOLOv8 comparison

Model	MAP	Inference time	Size
Faster RCNN	0.229	336.2 ms	~720 mbs
YOLOv8 nano	0.353	3.8 ms	~5.9 mbs