Google Vertex AI - Car Part Recognition

In this personal project, I worked with Google Cloud’s Vertex AI and AutoML Vision to develop a machine learning model for image classification. I started by uploading a labeled dataset of car part images to Cloud Storage and connecting it to Vertex AI as a Managed Dataset. After inspecting the images for errors, I initiated an AutoML Vision model training job to classify damaged car parts. Once the model was trained, I requested predictions from a hosted model using the same dataset to assess its performance.

To validate the model, I used TensorFlow tools such as TFRecords for efficient data loading, TFDV to generate statistics, and Facets for data visualisation. By analysing label distribution across subsets of the data, I was able to refine the model’s accuracy and further explore the predictions. This project allowed me to explore end-to-end model development and validation using Vertex AI and TensorFlow.