// Dataset
	In order to create an object detection algorithm that can identify litter, we needed a huge dataset that consisted of litter found on google streets.

	Finding litter by ourselves was no easy task. So to automate this time-consuming process, we ended up having to create script that could gathered thousands of random street images, whether they contained litter or not.

	The next step was preprocessing these images for training.

	This consisted of getting the human coders to use image labeling software to individually box and label all pieces of litter found in the street images that contained them. In total, over 3000 images were coded, with over 9000 pieces of litter labeled.

// Object Detection
	For our first prototype, our primary focus was getting tensorflow to identify all objects in an image. This was done by using an rcnn model that was pretrained to a dataset of common objects.

	Here, identified objects are boxed in green and along with its accuracy score or confidence level of the made prediction.

	On top shows the number of objects identified and a ranking assigned based off a set range of numbers where
		0 - 2 = rank 1
		3 - 5 = rank 2
		6 - 8 = rank 3
		9 + = rank 4

	How objects are identified using the rcnn model is a pretty complex process. The jist of how rcnn works is that
		First, it selectively searching an image for objects by making a bunch of region proposals where it groups together adjacent pixels by texture, color and intensity.

		Next, the features of these region proposals are computed and lastly passed into a linear regression model where the regions are classified as objects and then coordinates of boxes that surrounds the objects are tightened.

	For our second prototype, our focus was training our own rcnn model to specialize in identifying litter only using the litter dataset we gathered.

	Ill hand it back to andrew who will discuss the training process.