Image analysis with AWS Rekognition

By Al
Image analysis is one of those services with a very high WOW factor. Even if you take your time to study machine learning, deep learning and you are able to do your bits and pieces in this new brave world, your are still likely to get surprised by the capabilities. The big cloud providers analyze many millions of photos on a daily basis and the bigger the data set becomes the more predictive power it will gain.

At Pied Piper we will be mostly using these and other readily available capabilities for our team projects as opposed to build our own models. And of course we can do all this from the comfort of Python which we can easily integrate with other elements like Cloud Native Applications and Internet of Things for whole end-to-end solution.

AWS rekognition offers the following services:
  • Object and scene detection. It provides a confidence score for each object it detects
  • Facial analysis. Locates faces in an image and provides face attributes and emotions
  • Face comparison. It provides a similarity score
  • Facial recognition. Finds similar faces in a large collection of images
  • Celebrity recognition. Recognizes faces from a large collection of famous individuals
  • Image moderation. It identifies content deemed not appropriate for certain audiences, to make it easy for apps to filter it
I have played and provided Python code for the first 3 services which are the ones that interest me the most. However I have found that all three use very similar workflow, so I would expect the remaining 3 services will be very similar too. The code can be found in the following Github repository


For both the object and face detection use cases I have provided two different Python scripts: a simple script for a single photo, and another one to analyze a whole bucket. Use the second one with caution.

As in the previous post where we discussed the AWS Polly service, you are going to need:
  • AWS credentials (access key and secret key). The Github repository provides more details on how to make those accessible to the Python script
  • Boto3 module. Install it with "pip install boto3"
  • S3 Bucket with photos
Rekognition integrates very easily with S3. You can point the script to specific buckets or to specific photos sitting on S3 for them to analyzed. It can also integrate with Lambda (I will discuss Lambda in a future post) to for example trigger a function when a new photo is uploaded to a bucket so that Rekognition can tag it.

For more details about AWS Rekognition with Boto3 visit:

http://boto3.readthedocs.io/en/latest/reference/services/rekognition.html


Comments

Post a Comment

Popular posts from this blog

Sending PowerStore alerts via SNMP

By Al
Image
In this previous article we discussed how to leverage PowerStore's powerful API to collect audit logs and send them to a Syslog server . We used Logstash to do that because of its extensive list of plugins Sometimes I come across organizations that still have a requirement to send storage array Alerts to SNMP. As in the Syslog case, ultimately most modern monitoring tools today support collecting information via REST API because that's the way the application world is going, i.e. the cloud native way ... it would be fun to see Prometheus monitoring Kubernetes via SNMP :) However there is still some requirement out there for SNMP traps, so we will explore how to do it. For simplicity we will use the same toolset that we used in the previous article. I encourage you to revisit quickly the article as we did an extensive introduction to Logstash, with practical examples as well as a description of the logs available in PowerStore. I won't cover those aspects in this article I
Read more

Sending PowerStore logs to Syslog

By Al
Image
In this article we will explore how to get the logs from a DellEMC PowerStore array to a Syslog server. For this purpose we will use the PowerStore’s REST API, which is a great piece of engineering and a joy to work with as a developer. If you want to learn more about the PowerStore REST API I strongly recommend you quickly skim through the 2 articles I have written about the REST API. Part 1 - http://anzpiper.blogspot.com/2020/08/intro-to-dellemc-powerstore-rest-api.html Part 2 - http://anzpiper.blogspot.com/2020/08/powerstore-rest-api-best-practices.html In particular the second article demonstrates the capabilities of the REST API query language. This is a great feature I will use heavily in the last section so I strongly recommend you read that one at least As a side note, Syslog was developed in 1980. So this year it has turn 40 years old! That is a long time by any measure ... many of my colleagues were not even born in 1980. But in the technology scale it looks even scarier. 198
Read more

Electronic Nose - eNose

By Al
Image
In the previous article I provided details on VeeFarm which was one of the best projects we saw in ANZ Pied Piper 2019 edition and one of the best projects in the history of the program. The program inspired me and others to work together to showcase a great idea and the Pied Piper program itself during our presales conference. An idea I suggested to enhance VeeFarm was an electronic nose that could be used to detect the optimal ripening of the produce to advise the owner that it is the right time to pick it up the produce. This is an idea I had in the backburner for a while but never found the time to work on. Vee gave me the perfect excuse. The electronic nose, or eNose for short, is a custom PCB that hosts a range of gas sensors. For the proof of concept, I used the cheap MQ sensors. These are designed mostly for hobbyists and makers and they barely cost $1 per sensor. As one could expect, tweaking and calibrating these sensors it is even harder than the most expensive counter
Read more