Eliminate the time-consuming process of building and fine-tuning Deep Neural Networks (DNNs) from scratch for Intelligent Video Analytics (IVA) applications.

The term “transfer learning” implies that you can extract learned features from an existing neural network and transfer these learned features by transferring weight from an existing neural network. The pre-trained models accelerate the developer’s deep learning training process and eliminate higher costs associated with large scale data collection, labeling, and training models from scratch. Transfer Learning Toolkit enables you to build high performance IVA based applications such as retail analytics, logistics, smart cities, access control and more.

The Transfer Learning Toolkit is a python-based toolkit that enables developers to take advantage of NVIDIA’s pre-trained models and offers capabilities for developers to adapt popular network architectures and backbones to their own data, train, fine tune, prune and export for deployment. The simple interface and abstraction improves the efficiency of the deep learning training workflow.

Key Capabilities

  • GPU optimized pre trained weights for computer vision tasks
  • Easily modify configuration files for adding new classes and retraining models with custom data
  • Perform model adaptation and retraining in heterogeneous multi- GPU environments
  • Reduce model sizes using pruning functionality
  • Model Export API for deployment on NVIDIA DeepStream SDK with NVIDIA Tesla and Jetson products
  • Jupyter notebook examples for object classification and detection use cases

How-To Guide             Download             Documentation             Developer Forum            

Pre-trained IVA specific image classification and object detection models trained on selected public datasets are available to be used with Transfer Learning Toolkit.

    Image Classification

  • ResNet10/18/50
  • VGG16/19
  • MobileNet V1/V2
  • AlexNet
  • SqueezeNet
  • GoogLeNet

     

    Faster RCNN supporting backbones:

  • ResNet10/18/50
  • VGG16/19
  • GoogLeNet
  • MobileNet V1/V2

    Object Detection

    DetectNet_v2 supporting backbones:

  • ResNet10/18/50
  • VGG 16/19
  • GoogLeNet
  • MobileNet V1/V2

     

    SSD:

  • ResNet10/18

Enabling End to End Deep Learning Workflow for Intelligent Video Analytics

For developers designing and integrating IVA end applications such as parking management, securing critical infrastructure, retail analytics, logistics management and access control, etc. NVIDIA provides end to end deep learning workflow with DeepStream SDK for AI-based video and image understanding, as well as multi-sensor processing. DeepStream and Transfer Learning Toolkit together are an integral part of NVIDIA Metropolis, the platform for building end-to-end services and solutions for transforming pixels and sensor data to actionable insights.

Availability

  • Transfer Learning Toolkit is free and can be downloaded from NVIDIA NGC
  • Pre trained models can be downloaded for free to be used with the software
  • If you would like to use tlt-converter to run the downloaded models with NVIDIA DeepStream SDK and to convert model from UFF to a TensorRT engine. Download tlt-converter

Get Started With Hands-On Training

The NVIDIA Deep Learning Institute (DLI) offers hands-on training for developers, engineers and researchers in AI and accelerated computing. Sign up for a full day workshop focused on deep learning for IVA by contacting DLI directly.

Register Today

Tutorials & Technical Blogs

Webinars

Additional Resources