Image Captioning with Vision Encoder-Decoder

This project demonstrates an image captioning system that leverages a pre-trained vision encoder-decoder model to generate descriptive captions for images. It combines the power of computer vision and natural language processing using Hugging Face's nlpconnect/vit-gpt2-image-captioning model.

Overview

Image captioning is the task of automatically generating natural language descriptions for given images. This repository uses:

• Hugging Face Transformers to load a pre-trained vision encoder-decoder model.
• ViTFeatureExtractor to preprocess images appropriately.
• PyTorch as the deep learning framework.

The project is designed for quick experimentation and demonstration, and it runs on GPU if available.

Features

• Pre-trained Model: Uses nlpconnect/vit-gpt2-image-captioning for generating captions.
• Easy-to-Use: Simple function generate_caption(image_path) to produce captions.
• Flexible: Easily extendable for different image datasets or additional processing.