Labelbox•August 20, 2018
SomaDetect provides dairy farmers with the information they need to produce the highest quality milk with the resources they have today, for a more sustainable dairy food system. An important part of making this happen is is machine learning and computer vision.
We spoke with Bharath Sudarsan, Director of AI, who went into a little more detail about how SomaDetect is making a splash through the internet of cows.
“Our basic business goal is to make better milk. We have IoT devices that connect right into milking line with sensors scanning for metrics that are important to farmers. Our sensors can detect and communicate anything from protein and fat content, which are part of how milk is valued, to diagnostic tests that can reveal important indications about the health of the cows.”
“On big farms with 1,000 cows, it’s hard for farmers to keep track of everything. We’re bringing precision agriculture into dairy, and using AI with it.”
SomaDetect uses computer vision and deep neural networks to develop models based on the data they collect, which are used to predict and ensure milk quality and to rapidly diagnose health conditions, even before they occur, down to the individual cow. In addition, they are using this machine learning tech to expand their offerings to help farmers.
Employees wanted to join in on new AI projects being developed for SomaDetect, but were unable due to the open source tools they were using previously being too technically complex, and taking too much of Bharath’s team’s time to set up. “The barrier of entry for non-CS students who were willing to label was really high. I wanted something repeatable that would scale for the long-term.”
“Every few months, a new image annotation format comes out. With Labelbox, we have a professional group that focuses on creating and updating the data labeling platform to keep up with the latest technology. We as a company no longer need to worry about doing this ourselves; we can just rely on Labelbox.”
Creating and maintaining machine learning models requires a large amount of high-quality labeled data, produced especially for the type of tasks for which they will be used. “Labeling images can quickly become tedious and monotonous so people cut corners. A badly annotated image is way worse than no image at all. The fact that we have continuous monitoring with Labelbox is amazing.”
“The value we get from Labelbox is hours and dollars. Time I spend with someone’s laptop setting up an environment is two salaries being wasted. Now, I can explain to anyone how to do image segmentation in 5 minutes, tops.”
“All of our executives love it: we can put literally anyone on the data labeling project. Anyone can help.”
“[The Labelbox team] being so quick to respond has really impressed us. It was such a pleasant surprise. All we need to do is send a message, and they’re on it.”
“Working with Labelbox has increased my expectations for our AI projects because of the ability to create high-quality training data sets. Your model accuracy is directly proportional to how good your annotation is: the quality of the entire project improves.”
Labelbox•June 16, 2022
How Deque uses data prioritization and model diagnostics to unlock AI breakthroughs in digital accessibility
Deque has developed a sophisticated data engine that’s capable of prioritizing the most performant classes of data, discovering model errors quickly, and fueling their iterations with high-quality data.