Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Duration 14 hours
Course Outline
Introduction to CV/NLP Deployment with CANN
- The AI model lifecycle from training through to deployment.
- Critical performance factors for real-time CV and NLP applications.
- An overview of CANN SDK tools and their function in model integration.
Preparing CV and NLP Models
- Exporting models from PyTorch, TensorFlow, and MindSpore.
- Managing model inputs and outputs for image and text-based tasks.
- Utilizing ATC to transform models into OM format.
Deploying Inference Pipelines with AscendCL
- Executing CV/NLP inference via the AscendCL API.
- Preprocessing workflows: image resizing, tokenization, and normalization.
- Postprocessing tasks: bounding boxes, classification scores, and text generation.
Performance Optimization Techniques
- Profiling CV and NLP models utilizing CANN tools.
- Minimizing latency through mixed-precision and batch tuning strategies.
- Managing memory and compute resources for streaming tasks.
Computer Vision Use Cases
- Case study: Object detection for intelligent surveillance.
- Case study: Visual quality inspection in manufacturing processes.
- Creating live video analytics pipelines on Ascend 310.
NLP Use Cases
- Case study: Sentiment analysis and intent detection.
- Case study: Document classification and summarization.
- Real-time NLP integration with REST APIs and messaging systems.
Summary and Next Steps
Requirements
- Proficiency in deep learning concepts related to computer vision or NLP.
- Practical experience with Python and AI frameworks such as TensorFlow, PyTorch, or MindSpore.
- A foundational understanding of model deployment or inference workflows.
Target Audience
- Computer vision and NLP professionals working with Huawei’s Ascend platform
- Data scientists and AI engineers focused on developing real-time perception models
- Developers integrating CANN pipelines within manufacturing, surveillance, or media analytics domains
