This project is an AI-powered computer vision system that classifies store products (like sauce, noodles, and powder) from images or a live webcam feed. The goal is to automate product recognition using convolutional neural networks (CNNs), making it easier to track or identify items in a store setting.
Built with TensorFlow and OpenCV, the model is trained on a custom dataset of labeled product images. It uses MobileNetV2 as a foundational model that is retrained.
- Classifies products into categories
- Supports predictions on images or live webcam video
- Built using TensorFlow with MobileNetV2 for high accuracy
- Includes data preprocessing pipeline to clean and prepare input images
- Outputs predicted class and confidence scores
The dataset consists of:
- ~100 annotated images per class for training and validation
- Images were labeled using LabelImg and converted into a format compatible with TensorFlow pipelines.
- Feature extractor: MobileNetV2 (transfer learning)
- Classifier: Fully connected dense layers on top
- Trained using Adam optimizer, categorical crossentropy loss
- Early stopping and model checkpointing to avoid overfitting