This github repo is all about the course from Intel AI developer program, the course can be accessed here for registration https://software.seek.intel.com/DataCenter_to_Edge_REG
Using Intel hardware and sofware portfolio for demonstarting the Data Science process. Hands on understanding of building a Deep Learning model and deploying to the Edge.
- Using the concept of Transfer Learning for Image classification problem
- Exploratory Data analysis on Vehicle Make and Model Recognition (VMMR) dataset
- Training three different frameworks (Inception V3, VGG and MobileNet) on Intel Dev cloud
- Obtaining the graph and weights of the trained networks
- Evaluation of the models and freezing the graphs obtained
- Deployment to the Edge using Intel OpenVINO Toolkit
For Setting up the Environment on your workstation or on Intel AI Devcloud refer to Installation Instructions doc
The project starts with VMMR, contains 9170 classes, identified 76 Car Manufacturers and 291,752 samples in total. The time scale of the dataset covers models manufactured between 1950-2016. VMMRdb dataset contains images that were taken by different users, different imaging devices, and multiple view angles, ensuring a wide range of variations to account for various scenarios that could be encountered in a real-life scenario.
Data Creation, Preprocessing and Data Augmentation see Data_analysis
Training a model involves multiple steps, Choosing a framework (TensorFlow), Transfer Learning (InceptionV3, VGG16, MobileNet), Training the models and tune it for better performance, Hyper parameter tuning, Generate a frozen graph (a trained model) for inference on Edge.
Training and comparision of the performance & results see training
Interpreting the results of the training by analyzing the models with different metrics and graphs see here
- Confusion Matrix
- Classification Report
- Precision-Recall Plot
- ROC Plot
Inference on edge is real time evaluation of model subject to the contraints of power, latency and memory. Intel's OpenVINO toolkit optimizes pre-trained models for inference on intel processors.
-
Indentification of stolen cars
-
In same initialized environment, run the Inference_GUI.py script, you can see a working solution to the problem of creating a car theft classification application using Intel hardware and software tools.
-
In the GUI select the XML file generated by the Model optimizer, the labels file, test samples and also you can configure to run the inference on CPU, GPU or Movidius hardware.
For Deployment of the frozen model using OpenVINO Toolkit see deployment
-
Intel AI Academy https://software.intel.com/ai-academy
-
Intel AI Student Kit https://software.intel.com/ai-academy/students/kits/
-
Intel AI DevCloud https://software.intel.com/ai-academy/tools/devcloud
-
Intel AI Academy Support Community https://communities.intel.com/community/tech/intel-ai-academy