Скачать или смотреть Scene Captioning using Computer Vision and Deep learning Techniques

This video demonstrates our work on building an automated Scene Captioning system that bridges the gap between Visual Understanding and Natural Language Generation.

👥 Team Details :

Moturu Himesh (AM.AI.U4AID23013)

Besta Bhavya Sree (AM.AI.U4AID23033)

Akshay Reddy Velugati (AM.AI.U4AID23062)

Traditional image classification models identify objects but fail to describe relationships or context. Our project aims to generate accurate, human-like natural language descriptions for images. We implemented a pipeline using the BLIP-2 architecture, combining a Vision Transformer (ViT) for feature extraction with a Flan-T5 Large Language Model for text generation. The model was evaluated on the Flickr8k dataset.


🛠️ Technical Stack:

Architecture: BLIP-2 (ViT + Q-Former + Flan-T5)

Frameworks: PyTorch, Hugging Face Transformers

Dataset: Flickr8k (8,000 images, 40,000 captions)

Metrics: BLEU, ROUGE, METEOR

🤖 AI Tools Used for This Presentation:

Script & Content: Grok and Gemini

Audio/Voiceover: AI Text-to-Speech (Neural TTS).

Visual Avatar: Gemini pro and Hey gen ai

Video Editing: Canva

#ComputerVision #DeepLearning #AI #BLIP2 #SceneCaptioning #PyTorch