We were originally inspired by Val's idea to track animals outdoors. Although very interesting, it would have been very difficult to test for the purposes of this competition. We pivoted to brand recognition as that is easier to test. Many of the same visualization principles that we applied to brand recognition could easily be applied to a similar solution that detects animal types.
What it does
Our project uses the AWS DeepLens to detect certain brand logos. The data captured is sent to AWS where it can be visualized using AWS Quicksight.
How we built it
Challenges we ran into
None of us had very much image recognition or machine learning experience. It was challenging to figure out how to create a data set to train and also figure out how to finetune a pretrained model. Fortunately, the ssd example project in mxnet was an invaluable resource.
Accomplishments that we're proud of
Training a model that actually work!
What we learned
We gained great experience with a variety of AWS Services including, but not limited to SageMaker, Lambda, Greengrass, IoT, S3, Firehose, Quicksight, and CloudFormation. We also learned a lot different machine learning concepts, such as creating training data sets and finetuning an existing model. AWS makes it so much easier to learn about ML!