This is a web application built with Streamlit that classifies car models from images using a pretrained ResNet50 model. The model is fine-tuned on the Stanford Cars Dataset from Hugging Face, which includes 196 different car models.
Whether you're a car enthusiast or just curious about what model you're looking at, this app makes it easy to identify cars from photos.
- Upload a car image or enter an image URL
- Predicts car model with confidence percentage
- Shows a green confidence bar for easy visual interpretation
- Detects invalid images and reports "Not a car" if prediction confidence is low
- Responsive interface with background image support
- Backbone: ResNet50
- Dataset: Stanford Cars (Hugging Face)
- Number of classes: 196
- Prediction confidence threshold: < 20% returns "Not a car"