CANN SDK for Computer Vision and NLP Pipelines Training Course

Introduction to CV/NLP Deployment with CANN

• The AI model lifecycle, from initial training through final deployment
• Critical performance factors for real-time CV and NLP applications
• A comprehensive look at CANN SDK tools and their function in model integration

Preparation of CV and NLP Models

• Exporting models from frameworks such as PyTorch, TensorFlow, and MindSpore
• Managing model inputs and outputs for image-based and text-based tasks
• Leveraging ATC to transform models into the OM format

Implementing Inference Pipelines with AscendCL

• Executing CV/NLP inference through the AscendCL API
• Designing preprocessing stages: image resizing, tokenization, and normalization
• Handling postprocessing outputs: bounding boxes, classification scores, and text results

Performance Optimization Strategies

• Profiling CV and NLP model performance using CANN tooling
• Minimizing latency through mixed-precision computing and batch size adjustment
• Managing memory and computational resources for streaming workloads

Computer Vision Applications

• Case study: Implementing object detection for intelligent surveillance systems
• Case study: Visual quality control in manufacturing processes
• Developing live video analytics pipelines on Ascend 310 devices

NLP Applications

• Case study: Performing sentiment analysis and intent recognition
• Case study: Document categorization and summarization
• Integrating real-time NLP services with REST APIs and messaging systems

Conclusion and Recommended Next Steps