AI Model Compatibility
PAQi was designed to be AI model agnostic because we knew the power of AI would facilitate the rapid development of better models. For that reason, PAQi provides multiple methods for deploying inference (the means make predictions based on new images) on customer supplied hardware or the PAQi controller.
Model Types
YOLO
YOLO V8 is the current state of the art real-time object detection algorithm capable of detecting objects in an image with a high level of accuracy quickly, making it suitable for various manufacturing applications, including classification, part presence verification, and defect detection.
ResNet
ResNet is a deep convolutional neural network architecture known for its ability to train very deep networks effectively. It's widely used in image classification tasks and has achieved state-of-the-art performance on various benchmark datasets.
SSD
SSD is another real-time object detection algorithm that combines the speed of YOLO with the accuracy of region-based detectors like Faster R-CNN. It's particularly useful for applications requiring high detection accuracy and real-time performance, such as video surveillance and robotics.
Faster R-CNN
Faster R-CNN is a popular object detection framework that uses a two-stage architecture to localize objects in an image. It's known for its accuracy and has been widely adopted in applications like image recognition, face detection, and industrial inspection.
EfficientDet
EfficientDet is a family of object detection models that achieve state-of-the-art performance with significantly fewer parameters and faster inference times compared to previous architectures. It's known for its efficiency and scalability, making it suitable for a wide range of applications, including real-time object detection in industrial settings.settings. EfficientDet is the first of its kind because it is a open source model built by AI.
AutoML
The next wave of model development will be performed by machines. AutoML is the process of using machine learning to develop models based on the dataset type versus adapting a preexisting model to work with the dataset. AutoML is the future of deep learning and PAQi is prepared to serve AutoML models as the technology matures.
Model Training
The Platform
PAQi provides methods for deploying localized inference on trained models produced by multiple online platforms including but not limited to Roboflow, Vertex AI, and QC Hero.
Customization
Advanced users are able to train and deploy their own custom proprietary models in parallel to standard open-source trained models.
Localized Inference
By localizing inference on customer provided hardware or our PAQi controller, uptime is maximized by avoiding pitfalls associated with cloud-based inference such as phoning home. For applications requiring a high detector count, up to 30 object detectors can be run simultaneously using the QC Hero training platform.
Prediction Handling
Our patent pending calibration methods provide a quick no-coding means for turning predications into action whether it be signaling a robot to place a defect part in a reject bin, stopping an extruder from producing waste, or actuating a paddle to divert bad parts on a conveyor line.