Maasai Mara

This section describes all modules developed for the Maasai Mara team.

Objective

Primary objective of the project is to semi-automate the Maasai Mara's image processing pipelines using BisQue. In other words, the goal is to enable a Rear Seat Observer (at Mara Reserve) to upload the images to BisQue and semi-automate the image annotation and image analysis workflows by letting them leverage BisQue.

BisQue Usage Overview

Users can come to BisQue with a dataset of images and use BisQue as follows:

Predict and Export

• Upload the dataset to BisQue

• Run the AI model on the uploaded dataset

• Update the Predictions, if required

• Export the Annotations from BisQue

Annotate and Train

• Upload the dataset to BisQue

• Hand Annotate the Dataset or Import the Annotations to BisQue

• Train the AI model

Report Generation and Evaluations

• Upload the dataset to BisQue

• Run the AI model on the uploaded dataset

• Generate Reports on Animal Counts

• Compare the model predictions with ground truth (if exists)

• Build Dashboards using BisQue API

In other words, Users can just login to BisQue from a browser and do the following:

BisQue usage - Anticipated Flow chart

Developed Modules

The above mentioned flow can be achieved using the modules described below:

• Module to Tag Annotations from a text file

Mara Animal Annotator

• Module to Detect Animals using AI

Mara Animal Detector

• Module to Evaluate a machine learning model

Mara Evaluator

• Module to generate report on a given dataset

Mara Report Generator

• Module to Retrain the AI Model

Train a new detector