What does the problem look like?

A client comes in with an instance segmentation request. This means that they are engaged in or will be engaging in an activity where detection of a concept will be needed. This concept can be something found above water or underwater but they have the means to click pictures or take videos of that object in the environment in which it is found.

To qualify as this problem, the client must follow the below two pointers:

• They have at least one concept they are interested in segmenting
• They have a way to capture images/videos of that concept or they already have images/videos of that concept
• They have someone who can detect/segment that concept clearly in the dataset

Expected Output

From us, the client could expect an output like the following:

• A trained AI model that can be used to detect and segment target concepts from random images
• File containing segmentations for a group of test images that they want segmentations predicted for
• Images with segmentations drawn on them
• Other outputs depending on the above outputs

How can we help them?

We need the following:

(a) A set of images or a set of videos containing instances of that concept (jpg, png, tif)

(b) A set of labels on a subset of the above images or video frames

Post getting these, we should be able to run our code on the given data and share the output with them.

We refer to our code as a pipeline. This pipeline is simply a series of steps that are run on the images and annotations given by the client to .

Details on Individual Components of our Pipeline

Item Storage