VideoMarker: AWS Tools for Image Categorization
by Lis Cawley
ASR@ AACC --- This week Amazon Rekognition is up and running using the Amazon Web Service Command Language Interpreter. There are many labels in Amazon Rekognition. Amazon Rekognition does not allow an individual to create their own labels. However you can request your own labels and use Amazon Sagemaker to create a custom image classifier. [1]
Amazon SageMaker enables developers and scientists to build machine learning models for use in intelligent, predictive apps.[2] The Amazon SageMaker image classification algorithm is a supervised learning algorithm that takes an image as input and classifies it into one of multiple output categories. It uses a convolutional neural network (ResNet) that can be trained from scratch, or trained using transfer learning when a large number of training images are not available.[3] This week I may start extracting images using Media Analysis Solutions to prepare a data set that I can train Sagemaker to recognize muzzle flash. Muzzle flash is the visible light of a muzzle blast, which expels high-temperature, high-pressure gases from the muzzle of a firearm.[4] I will be researching to see if there are any already prepared muzzle flash data sets and also will be reading up on how to set up Sagemaker. Below is an image of the Uss Iowa Ship and a dramatic example of muzzle flash.[5] I will be recognizing much more subtle examples of muzzle flash from handheld firearms.
[2] https://aws.amazon.com/sagemaker/faqs/
[3] https://docs.aws.amazon.com/sagemaker/latest/dg/image-classification.html
[4] https://en.wikipedia.org/wiki/Muzzle_flash
[5] https://pixabay.com/en/uss-iowa-ship-us-naval-vessel-543589/
