The DeepView Model Pack provides developers with both public (Open Source) and production ready models.
Public Models as the name suggests, these are models that have been adapted, tested and validated and are ready for evaluation on common platforms from NXP and its partners using the eIQ Toolkit.
Production Models are custom implementations which have been fully optimized and validated on DeepView Starter Kits, NXP i.MX Application Processors and i.MX RT Crossover MCUs. These models include support and maintenance along with artifacts of provenance for all source material used during development.
Common public models that have been adapted, tested and validated and are ready for evaluation on common platforms from NXP and its partners using eIQ Portal.
MobileNet V1, V2, V3
MobileNet V1, V2, V3 SSD
ResNet V2 50
More coming soon
DeepView Commercial Models
Fine-tuned production-ready models where you decide the tradeoffs on performance, accuracy, and size. Available in floating-point and quantized implementations. These models can be fully tuned and re-trained, not limited to reference implementations.
The DeepView Commercial Models provide developers with quick way to maximize performance, take control of software lifecycle management issues and reduce or avoid the most common AI/ML project risks including IP Provenance & licensing liabilities.
DeepView object detection models deliver state-of-the-art detection for inanimate objects, people, or animals. Use standalone or with a MobileNet, ResNet, or custom backbone.
DeepView classification models are significantly more advanced than standard reference classification models such as ResNet or MobileNet. Models come pre-trained with high-level features with support for transfer learning to finish training with your dataset.
Pose and Gesture Models
This example showcases the PoseNet model running on the DeepViewRT inference engine to provide a very efficient Pose and Gesture recognition solution. The demo shows an application built using QML that can detect and overlay an outline of a person or persons’ joints and limbs onto a video feed using a PoseNet model.
DeepView segmentation models generate a pixel-wise mask of the inanimate objects, people, or animals and can be used in conjunction with other DeepView models to perform tracking, identification, and reidentification.
License Plate Recognition (LPR/ANPR)
DeepView license plate recognition models providing an intelligent way to improve your tracking, identification and reidentification to help train and enhance your dataset.