During our recent Image Processing release, we shared that it sets the foundation for more image data-driven capabilities on the BigML platform. As promised, we are launching Object Detection on Tuesday, June 28, 2022, at 08:00 AM PDT / 05:00 PM CEST, with a FREE live webinar. In this post, we make a quick introduction to the basic concepts involved before we move on to the remainder of our blog post series that will give you a detailed understanding of all things Object Detection: annotations, workflow automation, use case examples, and a peek into how BigML’s implementation works under the hood.
Object Detection Defined
Object Detection is a fundamental computer vision task that deals with detecting instances of semantic objects of predefined classes (e.g., human beings, products in a store, vehicles in traffic) in digital images and videos. Object detection models determine the objects that appear in a given image along with the exact coordinates for each identified object in the form of bounding boxes. This differs from the previously launched Image Classification, which aims to assign a label (or labels) to a given image from among a predefined set of class labels, e.g., truck, motorcycle, SUV, or car.
The part of the process that involves finding the correct regions in the image is known as localization. This can get tricky for images where multiple objects of interest have overlapping regions. For example, if a person is riding a bicycle the bounding boxes for each naturally overlap. Luckily, modern Object Detection techniques are advanced enough to handle such situations and deliver actionable insights from your image data.
BigML’s Object Detection utilizes a specialized neural network known as YOLO (You Only Look Once) that pre-processes a given image in a single run by analyzing equally partitioned regions and assigning each region a match score per object class being predicted. If there are overlapping regions, those can be considered a match or not by applying user-defined thresholds. This approach has become very popular as it is proven to drastically speed up performance while not sacrificing much in the accuracy of the detections.
Object Detection Use Cases
Almost any business these days has access to valuable image data that remain underutilized from an insight generation or inference-making perspective. BigML Image Processing is designed and architected to democratize and accelerate the launch of image-based smart applications. Below is a small sample of real-world applications that leverage image data and deliver tangible benefits.
- Manufacturing: There are many productive uses of Object Detection in manufacturing. Quality control and defect detection is one such example that helps process engineers take corrective action upstream instead of downstream in the manufacturing process, where it becomes much more costly to address defects by introducing various exceptions.
- Retail: Object Detection presents lots of opportunities in retail too. Analyzing store video feeds to constantly monitor shelves helps automatically identify out-of-stock situations resulting in lost sales. This is a great way to protect margins in what typically is a low-margin business dependent on moving lots of products in volumes.
- Agriculture: In agriculture and animal husbandry, valuable livestock can be counted and monitored to ensure their safety and even for early detection of health risks based on their movement patterns.
- Healthcare: Disease diagnostic is an excellent fit for Object Detection. The example above is a mammogram x-ray that can be analyzed for different types of irregularities like benign or malignant tumors. State-of-the-art Object Detection systems have already matched and even exceed human expert performance in some of those tasks supporting better patient outcomes.
- Sports: Professional franchises and collegiate sports teams have rich image datasets that can be fed to deep learning models to analyze player positions and in-game performance which in turn become a great decision-making tool for coaches and managers who are increasingly taking the much-touted “Moneyball” approach to their profession.
- Transportation: Different types of vehicles traveling on city roads and highways and their license plates are ideal targets for Object Detection systems. For instance, at BigML, we have first-hand experience in delivering license plate recognition solutions that enable streamlined billing for toll road concessions with managed lanes.
What makes BigML Object Detection different?
By virtue of the fact that Object Detection is now available as a native feature on our platform, it inherits all that makes BigML Image Processing unique and a breeze to use as summarized below:
All the Tasks for Generating Insights from Image Data on a Single Platform
From labeling to inference, evaluation, and predictions all the relevant tasks for generating insights from your image data and automating business processes can be performed in the same tool. Since BigML treats images as any other data type on the platform, BigML users can easily use image data alongside text, categorical, numeric, date-time, and item data types as input to any type of Machine Learning model, both supervised and unsupervised. No more installing, maintaining, and jumping from one specialized tool to another.
Streamlined Image Dataset Management with Composite Sources
A composite source is a collection of component sources supporting multiple data types including individual images. Composite sources save time and prevent errors because users can incrementally add more data to them and use the built-in image labeling while preserving immutability along the way.
Comprehensive Feature Extraction Options to Feed Any Algorithm
Through multiple feature extraction configuration options, BigML gives you fine control over what the algorithm “sees” in your image data. As part of pre-processing, a collection of images are automatically transformed into rows of tabular data that any model type can handle with ease.
Pre-Trained CNNs for Classification and Regression
BigML also allows the user to pick from different pre-trained Convolutional Neural Nets to build better image processing models faster by leveraging Transfer Learning. These pre-trained models are based on industry-standard datasets containing millions of images and thousands of classes.
Want to know more about Object Detection?
If you’d like to learn more about how Object Detection works, please visit the release page as it will be continuously updated to include links to the upcoming blog posts, the latest BigML Dashboard and API documentation, the webinar slideshow as well as the full webinar recording past the official launch date. One last thing, register today for the FREE live webinar on Tuesday, June 28 at 8:00 AM PDT / 05:00 PM PDT to save your spot!