Case Study: EyeControl boosts eye-tracking accuracy with AllCloud’s ML pipeline

EyeControl boosts eye-tracking accuracy to 98.58% with AllCloud

EyeControl, a company providing a wearable eye-tracking communication device for patients, faced the challenge of improving the accuracy and inference time of its existing machine learning model. Their goal was to replace a 92% accurate image classification system with a more complex, two-step pipeline involving object detection and a classifier, all while managing project costs and timelines. They partnered with vendor AllCloud to develop this complete ML pipeline.

AllCloud developed a solution using the SSD algorithm for object detection and a confidential classifier, significantly reducing training costs through the use of AWS Spot Instances. This approach allowed for effective tuning and many iterations. The new pipeline implemented by AllCloud achieved a 98.58% accuracy rate on the test set with an average inference time of 266ms, vastly outperforming the customer's legacy model.

EyeControl