? SSL Salesforce Einstein Vision

Shen Yifeng

salesforce enthusiast,sfdc developer, consultant

Salesforce Einstein Series Round 3 : Vision

Einstein Vision

Nowadays, AI as a service is popular and Salesforce is catching up. Its aritifical intelligence products cover a wide range of areas including NLP, Vision, Voice etc...... This post focuses on Salesforce Vision, which is a powerful tool to find images' contextual information without one's professional machine learning knowledge. In real world scenarios, they can be applied in (and not limited to) :

  • Identifying customers’ preferences
  • Customer signature prediction
  • Quality of products
  • Visual search
  • Brand detection
  • Inventory levels

They are so meaningful and wonderful. As said in this official blog, take the visual search for example, it can expand the ways that customers can discover company's products and increase sales. Many more use cases we could imagine in this commercial world.

Until now, the image recognition functionality provided by salesforce mainly be categorized into two types: object detection and image classification. Here, I show two quick demos to see the interior methodologies:

1. Object Detection

Salesforce provides four types of image classifier : General Image Classifier, Food Image Model, Scene Image Model and Multi-Label Image Model. I implemented the API and deployed it on Heroku using django.

Let's see the following pictures which showed the prediction results.

And put another similar image into test:

The API document link is here :metamind.readme.io

The above sample may be used in brand detection, the next one is more universal since image search requirements can be seen everywhere and it requires human interaction(supervised learning).

2. Image Classification

Or reffered to as Image-Based Search, was first introduced in dreamforce at 2017 by showing the well-known demo called Dream House. Its reference link is here: www.dreamhouseapp.io Here is the quick deploy link : developer.salesforce.com/dreamhouse/einstein-vision . On my dev org, the dataset include three groups, every group has 16 traing pictures. They are labeled as Victorian, Contemporary and Colonial.

Below is the prediction result when a new house image put into analyze.

Evolution of AI

As shown in the below, artificial intelligence technology is always evolving, I am looking forwad to see salesforce's next generation deep learning vision products.

“AI is the goal; AI is the planet we’re headed to. Machine learning is the rocket that’s going to get us there. And Big Data is the fuel.” – Pedro Domingos

This post is a quick summary of Salesforce Einstein Vision.