AI developers, data scientists and companies building intelligent video analytics apps face significant challenges in creating and deploying highly accurate AI. Some of the key issues include dataset collection and labeling, achieving high accuracy with available dataset, deploying on legacy infrastructure, and scalability of apps and services.
With billions of cameras and sensors deployed across the world, the groundwork for Intelligent Video Analytics has already been laid out. Vision AI across industries has had a tremendous impact on business growth, driving operational efficiency and reducing cost. For instance, the technology has aided with traffic congestion, generating customer heat-maps at retail stores, and patient monitoring at hospitals and more.
To address increased demand, NVIDIA launched new highly accurate models and enhancements to AI software, targeting NVIDIA datacenter GPUs and Jetson platforms.
The pre-trained models and Transfer Learning Toolkit help data scientists, computer vision developers and software partners accelerate AI training. Developers can build highly accurate AI for several popular use cases using purpose-built models. NVIDIA DeepStream SDK helps developers and companies to build performant vision AI apps and services that can be deployed at scale and managed with ease using Kubernetes and Helm Charts.
Major Feature Enhancements:
Transfer Learning Toolkit 2.0 (developer preview)
The Transfer Learning Toolkit simplifies computer vision tasks and eliminates the time-consuming process of training from scratch. It creates highly accurate AI models for deployment with limited AI expertise.
The developer preview release extends training support on several popular and state-of-the-art networks to achieve greater inference throughput. The pre-trained models and TLT container are freely available and can be downloaded from NGC. Key highlights include:
- New NVIDIA purpose-built models: Deploy high fidelity AI applications for common use cases such as people counting, vehicle tracking and heatmap generation using highly accurate purpose-built models.
- DashCamNet
- Facedetect-IR
- PeopleNet
- TrafficCamNet
- VehicleMakeNet
- VehicleTypeNet
End to end performance for purpose-built models from TLT for deployment with DeepStream
Tabulated data is in FPS with 1080p input
† Inference at FP16 on Jetson Nano
Running on the DLAs for AGX Xavier and NX frees up GPU for other tasks
- Extended support for object detection models such as YOLO-V3, SSD, and FasterRCNN, RetinaNet, DSSD and DetectNet_v2
- Out of the box compatibility with DeepStream SDK 5.0 developer preview makes it easy to deploy models from TLT on the edge
- Speed-up AI model training with multi-GPU support
GETTING STARTED
- Download Transfer Learning Toolkit
- Check out the developer tutorial on how to use purpose-built pre-trained models with Transfer Learning Toolkit >> Developer tutorial
DeepStream SDK 5.0 (developer preview)
DeepStream is a streaming analytics toolkit for AI-based image and video understanding. By using DeepStream, you can build efficient edge applications to perform real-time AI.
The developer preview release introduces enhancements and features for system designers, AI developers, software partners and OEMs building IVA apps and services across many industries including smart cities, retail analytics, health and safety, sports analytics, robotics, manufacturing, logistics and more
- Run popular Deep Learning framework natively with DeepStream: New inference capability with Triton Inference Server (previously TensorRT Inference Server) enables developers to deploy a model natively in TensorFlow, TensorFlow-TensorRT, PyTorch, or ONNX in the DeepStream pipeline
- Python based development: Select from either C/C++ or Python based applications for your use-case. We’ve introduced Python support with seven reference apps <link>. Both development options provide comparable performance
- Secure communication between edge and cloud using Kafka message broker and TLT authentication
- IoT capabilities
- Superior DeepStream app control from edge or cloud with bi-directional IoT messaging
- Dynamic AI model update on the go reduces app downtime
- Interoperability with Transfer Learning Toolkit 2.0
- Introducing Jetson Xavier NX support: Deploy AI applications on the world’s smallest AI supercomputer at the edge. Reference apps are common among all platforms.
GETTING STARTED
- Download DeepStream SDK
- Check out the developer preview feature explainer blog >> Developer blog
- View the free self-paced online DeepStream DLI course on Getting Started with DeepStream for Video Analytics on Jetson Nano
Sign up for the NVIDIA developer program to stay in the know about upcoming IVA and IoT announcements, software releases, contests, projects and more.
