Building end-to-end computer vision solutions from deep learning models
summary
Microsoft shows how to turn ImageNet-scale research into usable pipelines using the Microsoft Cognitive Toolkit, Custom Vision Service and Azure Machine Learning Package for Computer Vision. The workflow uses transfer learning on pre-trained ResNet-style nets (ImageNet >10M images, ~22k labels) to get 93–94% initial accuracy on a deer classification test and 98–99% after adding hard negatives via active learning. Models train in minutes, export to ONNX, deploy via REST to Azure or edge, and support image classification, detection and similarity.
AI-written summary · about
chapters (12)
- 0:00Start
- 2:13Image Classification Use Case
- 2:41Object Detection
- 3:52Feature Descriptor
- 6:24Convolutional Layers
- 11:32Transfer Learning
- 13:28How Can We Use Pre Trained Deep Neural Nets
- 18:01Custom Vision Service
- 19:32Create a Project
- 22:29Microsoft Cognitive Toolkit
- 23:42Resources
- 33:05Certificate of Appreciation
related talks
16mHow to get started with Computer Vision AutoML in Azure
58mDeep Learning in Microsoft Azure: CNTK, CaffeOnSpark and Tensorflow
59mThe bold promise of teaching computers to understand the visual world
41mAI + Machine Learning for .NET Developers
01h 00m5 Levels of AI in Microsoft Azure
26mIntroduction to the Microsoft Cognitive Toolkit (Deep Learning Framework)
01h 02mAI on the Edge
29mApplied Custom Vision