Accelerating Vision AI Applications Using NVIDIA Transfer Learning Toolkit and Pre-Trained Models

Discover how to accelerate the creation of vision AI models using NVIDIA Transfer Learning Toolkit and pre-trained models in this 47-minute video. Explore the wide adoption of vision AI across industries and learn how to overcome the challenges of training accurate and performant deep learning models. Dive into the Transfer Learning Toolkit, a simplified AI toolkit for developers to train models without coding. Examine pre-trained models available on NGC and learn how to fine-tune them for specific use cases. Explore techniques like model pruning and INT8 quantization to build high-performance models for inference, reducing development effort and speeding time to market. Cover topics such as quantization-aware training, automatic mixed precision, instance segmentation, and specific models like PeopleNet and Face Mask Detection. Walk through the training workflow, including data preparation, model specification, training, evaluation, and deployment using DeepStream.

Intro
TRAINING CHALLENGES
TRANSFER LEARNING TOOLKIT (TLT)
TRANSFER LEARNING TOOLKIT 2.0
PURPOSE BUILT PRE-TRAINED NETWORKS Highly Accurate Re-Trainable Out of Box Deployment
QUANTIZATION AWARE TRAINING Maintain comparable Performance & Sperdup Inference using INTB Precision
AUTOMATIC MIXED PRECISION (AMP) Train with half-precision while maintaining network accuracy same as single precision
INSTANCE SEGMENTATION - MASK R-CNN
PEOPLENET
FACE MASK DETECTION
TRAINING WORKFLOW
CONVERT TO KITTI
TLT SPEC FILES
PREPARE THE DATASET
TRAIN - PRUNE - EVALUATE
TRAINING SPEC - DATASET AND MODEL
EVALUATION SPEC
TRAINING & EVALUATION
MODEL PRUNING
RE-TRAIN & EVALUATE
TRAINING KPI
QUANTIZATION & EXPORT
INFERENCE SPEC
DEPLOY USING DEEPSTREAM
SUMMARY