Image classification is a machine learning technique that automatically classifies images into one or more predefined categories. It is commonly used in a variety of business scenarios, including but not limited to product categorization, quality control, and security and surveillance.
One company that found success with image classification was a large pharmaceutical corporation. The company was responsible for manufacturing and distributing a wide range of medication. They wanted to ensure the quality and consistency of their products. They used image classification to create a model that could accurately classify their capsules and tablets into specific categories, such as dosage, shape, and colour. With its inventory correctly classified, the company could improve its quality control processes and ensure the consistency of its products. They now quickly identify any potential issues or abnormalities in their medication, which allows them to take appropriate action.
There are several platforms and libraries out there that help you classify images with a couple of clicks. Still, there are some factors you have to consider when:
- The accuracy of such a system is of utmost importance to you
- You have a large amount of data.
The lack of transparency and flexibility blinds users into believing that this is the best they can do, which isn't the case in most situations. This blog will show you how we built an image classification model in less than an hour and got State-of-the-art results.
Data Exploration
In this tutorial, we will look at a dataset known as the Chest X-Ray Dataset, in which input data is in the form of an image and target labels are “Normal” and “Pneumonia”.
The size of the dataset is around 2.5 GB with about 5200 Images.
MLOps on Simplismart
In this section, we will go through different phases like uploading a dataset, training an image classification, and evaluating and predicting by deploying the model on Inference Server. Before heading t the tutorial, we suggest you read Introduction to SimpliSmart Platform and sign up so you can follow us along in the tutorial.
Uploading Dataset
- Let's start with uploading the dataset on SimpliSmart. Head over to Dataset Tab from the sidebar menu, click on Add Dataset, and fill in details like Name of Dataset and Description (Optional)
- Click on Next, choose the source of the data frame, whether to upload from local storage or import from cloud storage. We will upload the d ta frame from local storage for this blog.
- Click on Upload a Dataframe; a dialogue box will open; choose the data frame, click on Upload Media files and upload the zip format image folder, click finish, and wait until it is uploaded.
- Once it gets uploaded, wait for a maximum of 5 minutes to get all details of the dataset, like the size on disk and the snapshot of the dataset like below.
Model Training
- Click on Train Model, Add basic details like Name of Model and Description and click Next.
- Select the dataset on which you want to train the model, select the output and input columns, and click next.
- Select the Number of Experiments, Time Limit and Hardware to train your model and click Finish.
- You would see the job as running in the card. Click the card and get ore details like configuration, experiments and metrics.
- As we can see from the above metrics, we have achieved an accuracy of 98.7% in train metrics and 96.5% in test metrics in 40 minutes with hyper optimization parameter tuning.
- Head to the Graphs section and turn some variables on to see the graph between different entities.
- After it flags success, head to the model's section and click on deploy. Once deployed, it will provide a Live Flag and click on the card.
- Click on Predict and upload an image from your local. Scroll down and click o predict.
Final Thoughts
So you see how quickly we have done 5 processes of the modern MLOps lifecycle without any code that provides SoTA results.
Image classification is a powerful tool that can help businesses quickly and accurately analyze and understand large volumes of visual data. With the right tools an resources, even people with little or no coding experience can build their own image classifiers to fit their specific needs.
When choosing a no-code AI platform to build an image classifier, there are several factors that you should consider about the platform:
Capabilities: The platform should have the tools and features to support image classification, such as pre-built models, algorithms, and customization options.
Ease of use: The platform should be user-friendly and easy to use, even for people with little or no coding experience. This will make it easie to create and train your image classifier without extensive technical knowledge.
Support and resources: The platform should provide access to a range of support and resources, such as tutorials, documentation, and a community of users, to help you get started and overcome any challenges you may encounter.
Pricing: The platform should offer flexible pricing and licensing options that fit your budget and requirements. It should also provide lear and transparent information about the costs and limitations of different pricing plans.
Security and privacy: The platform should have robust security and privacy measures to protect your data and models. This is particularly im ortant if you are dealing with sensitive or confidential information.
