Assurance Car Damage Assessment

📜 Presentation

These project is another version of my previous project Car Damage Toolkit with new awesome features :

• A pretty Interface powered by vuejs and Bulma CSS

---

• Backend handled by Django Rest Framework to facilitate API building

---

• Axios to facilitate Requests on the API

---

• Openvino Toolkit to transform each Tensorflow model into an Intermediate Represention to allow a fast Inference on CPU
  
  

---

• Possibility to generate a PDF report from Analysis's Result by using html2pdf.js

⏳ Workflow

The Analysis of each image follow these steps :

1. VGG19 model is used to determine if Image is a Car or Not
2. A second model is used to see if there is any damage or Not
3. The aim of the third model is to localize damage following three classes : Front, Rear, Side
4. Finally the last model determine the severity of damages based on three classes : Minor, Moderate, Severe

Demo

📆 Futur Works

• After the conversion of VGG19 model into an IR, it's performance has degraded too much. So actually the application just use the simple VGG19 during the Car or Not Checking. Because of these I want to implemente Post-Training Optimization with Openvino for these model
• Re-do the implementation of the PDF report to capture all the details on the page
• Dockerize the django-vue application