LIVE IN: https://09cf85aa05a949d1cd.gradio.live

Image Captioning with Vision Encoder-Decoder Model

An interactive Gradio application that generates captions for images using a pre-trained Vision Encoder-Decoder model (ViT-GPT2).

Features

• 🖼️ Generate captions for uploaded images or image URLs
• ⚡ Optimized for both CPU and GPU (automatically detects available hardware)
• ⚙️ Adjustable generation parameters (max length, number of beams)
• 🌐 Works with local images and web URLs

Technical Details

Model Architecture:

• Encoder: Vision Transformer (ViT)
• Decoder: GPT-2 language model

Pretrained Model: (https://huggingface.co/nlpconnect/vit-gpt2-image-captioning)

Installation

git clone https://github.com/yourusername/image-captioning-app.git cd image-captioning-app

The application will launch a local web interface accessible at http://127.0.0.1:7860

How to Use

Upload an image or paste an image URL Adjust generation parameters if needed (in Advanced Settings) View the generated caption Try different images to see various results